Lightning triggers ... nuclear reactions

ENVIRONNEMENT The power of thunderstorms is sometimes compared to that of the H bombs. This parallel makes perfect sense since physicists have just shown that they are the seat of nuclear reactions.
DENIS DELBECQ Facebook @effetsdeterre

Twenty-five years after the first observation of flushes of gamma radiation of terrestrial origin, quickly associated with storms, doubt is no longer allowed: lightning generates nuclear reactions, like our particle accelerators or stars.

A group of Japanese scientists, led by Teruaki Enoto (Kyoto University) describes his discovery this Thursday in Nature. "We installed detectors on a site on the east coast of Japan in 2006," says the researcher. In this region, thunderstorms are frequent and their clouds are rather low, which increases our chances of seeing what happens there." On February 6, the instruments delivered a clear signal for almost a minute, much longer than the cloud gamma flashes, which last only a few fractions of a millisecond.

The flashes of terrestrial origin
The gamma ray represents the most energetic form of the light spectrum. It is associated with nuclear reactions in stars, but also with uranium reactors and particle accelerators. Twenty-five years ago, an American space telescope freshly put into orbit to observe cosmic gamma rays, began to see, by chance, brief flashes of terrestrial origin, the TGFs. An invisible radiation, not to be confused with the mysterious sprites, clearly visible those at the top of certain storms, that we also seek to explain.

"The many observations made it possible to understand that TGFs are formed at the very heart of thunderstorms", says Sébastien Célestin, from the University of Orléans' Laboratory of Physics and Chemistry of the Environment and Space, involved in the one of the instruments of the French Taranis satellite - named after the Celtic god of the storm - to be launched in 2019 to detect particles and radiation from thunderstorms. "While we originally thought these very rare flashes, it would happen at least a thousand times a day in the Earth's atmosphere."

Antimatter production
When lightning occurs, it forges, on its way, a plasma. A soup of charged particles, made of nuclei of atoms and electrons. These particles emit all kinds of radiation, starting with the fairy lights that we see from our windows. "The electron energy of this plasma is initially about 1 electronvolt (the unit of measure used for particles), adds Sébastien Célestin. But TGFs, the gamma radiation they emit when they encounter matter, can reach tens of millions of eV. The electrons have therefore undergone a phenomenal acceleration that we are trying to understand." In other words, storms are real particle accelerators, synchrotrons.

In 2011, physicists made another surprising observation: thunderstorms produce antimatter. It is more specifically the positron, the antiparticle of the electron, which has the same mass as this one, as well as the same electric charge but of opposite sign. The meeting of the two annihilates them by emitting a light whose energy - 511 keV - corresponds very exactly to the mass of the pair of particles, by virtue of the famous relation E = mc2 of Einstein.

Chain reactions
In other words, TGFs initially produce electrons and positrons - by transformation of gamma radiation following the reverse reaction of annihilation - which are then annihilated by emitting gamma radiation betraying their past presence.

If this light at 511 keV is well observed, it is not alone, a sign that other mechanisms are at work. Physicists have over the years acquired the conviction that they were induced nuclear reactions which alone could explain the long duration of certain signals as well as the presence of free neutrons in thunderstorms. Remained to demonstrate, what did the Japanese group.

"This work is rigorous and the measurements are in agreement with the simulations. This result will make noise! "Rejoices Sébastien Célestin. According to Teruaki Enoto, "it is initially a photo-nuclear reaction, the absorption of gamma radiation by the nitrogen nuclei present in the air." A reaction that produces, in particular, positrons and neutrons. The latter generate a series of chain reactions that explain the long duration of Japanese observations, which in particular create carbon 14.

"It will explain the acceleration of electrons that is the source of TGF. This may tell us why all storms do not create such flashes, "said Sebastien Celestin. For its part, Teruaki Enoto hopes to "determine the share of atmospheric carbon 14 associated with storms. According to our initial calculations, it would be marginal. "

Thanks to a group of Japanese scientists, it is now attested that lightning
generates nuclear reactions, like particle accelerators or stars,
(WILD HORIZONS/UIG VIA GETTY IMAGES)

Lightning makes new isotopes

Physicists show that thunderstorms trigger nuclear reactions in the atmosphere.

DAVIDE CASTELASTELVECCHI

A streak of lightning in the skies over Japan has generated positrons - the antimatter equivalents of electrons - and radioactive carbon-14, confirming a theoretical prediction, according to a paper published in Nature on 22 November.

Since the 1990s, orbiting observatories designed to observe the heavens have also detected flashes of γ-rays coming from Earth, which were thought to have their origins in atmospheric phenomena. To investigate this theory, Teruaki Enoto, an astrophysicist at Kyoto University in Japan, and his collaborators set up an array of γ-ray detectors close to the Kashiwazaki-Kariwa nuclear power plant. Winter thunderstorms in Japan are famous for their spectacular lightning, he says, and the low clouds make these relatively easy to observe.

On 6 February, the detectors sensed an unusual event. A double lightning bolt just off the coast shot out an initial, one-millisecond spike of γ-rays, with relatively high energies of up to 10 megaelectronvolts. This was followed by a γ-ray afterglow of less than half a second. Then there was a telltale signal - γ-rays concentrated at 511 kiloelectronvolts of energy, which lasted for about a minute. Physicists say this is the unmistakable signature of positrons annihilating in a puff of energy as they hit electrons in the surrounding matter.

Together, the three waves of γ-rays point to a photonuclear reaction first proposed[2] a decade ago by Leonid Babich, a physicist at the Russian Federal Nuclear Center in Sarov. Lightning can accelerate some electrons to almost the speed of light, and the electrons can then produce γ-rays. Babich proposed that when one of these γ-rays hits the nucleus of a nitrogen atom in the atmosphere, the collision can dislodge a neutron. After briefly bouncing around, most of the neutrons get absorbed by another nitrogen nucleus. This adds energy to the receiving nucleus and puts it in an excited state. As the receiving nucleus relaxes to its original state, it emits another γ-ray - contributing to the giveaway γ-ray glow.

Meanwhile, the nitrogen nucleus that has lost one neutron is extremely unstable. It decays radioactively over the next minute or so; in so doing, it emits a positron, which almost immediately annihilates with an electron, producing two 511-keV photons. This was the third signal, Enoto says. He suspects that his detectors were able to see it only because the briefly radioactive cloud was low, and moving towards the detectors. This combination of circumstances might help to explain why the photonuclear signature has been seen so rarely. Enoto says that his team has observed a few similar events, but that the one described in the paper is the only clear-cut event so far.

Babich also predicted that not all of the neutrons dislodged from nitrogen by a γ-ray are absorbed. Some of them instead will trigger the transmutation of another nitrogen nucleus into carbon-14, a radioactive isotope that has two more neutrons than ordinary carbon. This isotope can be absorbed by organisms; it then decays at a predictable rate long after the organism's death, which makes it a useful clock for archaeologists.

The main source of the carbon-14 in the atmosphere has generally been considered to be cosmic rays. In principle, lightning could also contribute to the supply. But it is not clear yet how much of the isotope is produced in this way, says Enoto, in part because it's possible that not all bolts initiate photonuclear reactions.

"I agree with their interpretation of their data," says physicist Joseph Dwyer of the University of New Hampshire in Durham. But, he adds, Enoto's team's explanation does not solve all puzzles related to positrons in the atmosphere. In particular, the photonuclear reaction does not seem to match an event Dwyer observed in 2009 from a research aeroplane. His detector spotted a signature of positrons only for a fraction of a second - too short to originate from nuclear decay, he says. Also, his detector saw no initial flash in that case. "If it was there, it should have been very obvious."

1. Enoto, T. et al. Nature http://dx.doi.org/10.1038/nature24630 (2017).
2. Babich, L. P. JETP Lett. 84, 285-288 (2006).

Thunderous nuclear reactions

The discovery that thunderstorms can trigger nuclear reactions provides insight into the physics of atmospheric electricity and unveils a previously unknown natural source of radioactive isotopes on Earth. See Letter p.481

LEONID BABICH

Thunderstorms are some of nature's most spectacular phenomena. Almost a century ago, it was suggested that the strong electric fields in thunderclouds could accelerate electrons in the atmosphere and induce nuclear reactions[1]. However, these processes have been difficult to confirm experimentally. On page 481, Enoto et al.[2] report the first conclusive observational evidence for thunderstorm-produced nuclear reactions - with implications for our understanding of Earth's atmosphere and isotopic composition.

The idea that thunderstorms can trigger nuclear reactions was proposed[1] by the Scottish physicist and meteorologist Charles Wilson in 1925. However, the state of physics at the time meant that Wilson could not fully substantiate his idea. For instance, it is now known that neutrons are among the possible products of nuclear reactions and therefore that detecting these particles from a thunderstorm would provide evidence for Wilson's proposal. But neutrons were not discovered[3] until 1932.

Thunderstorms occur in the dense lowermost layers of the atmosphere. Electrons in these layers undergo frequent collisions with air molecules and are therefore subject to a strong drag force. Wilson's proposal requires electrons that have sufficiently high initial energies to overcome this force. It is now known that cosmic rays irradiate the atmosphere and produce such electrons, which multiply in thunderclouds to form an avalanche of highenergy electrons[4]. However, in the mid-1920s, cosmic rays were extremely mysterious and thought to be of terrestrial origin[5].

The first claimed detection of neutrons from a thunderstorm was reported[6] in 1985. These observations were carried out in the Himalayas in a region that has extremely high thunderstorm activity (about 30 lightning strokes per day). Since the late 1990s, many other studies have also claimed statistically significant detections of thunderstorm-produced neutrons from all over the world[7-10]. However, the detectors could not distinguish neutrons from other particles such as electrons and γ-ray photons - all three would produce similar electric-current pulses in the detectors[11].

It was initially thought that thunderstorm-induced neutrons were produced in a nuclear reaction in which two nuclei of the hydrogen isotope deuterium fuse in the plasma created by lightning to form a helium nucleus and a neutron. However, it was later shown that the physical conditions in such a plasma do not allow this reaction to occur[12].

Instead, the avalanche of high-energy electrons produced in a thundercloud emits X-ray and γ-ray photons. Since the late 1980s, these photons have been detected on the ground, by aircraft flying inside thunderclouds, and by artificial satellites in near space (about 500 kilometres above Earth's surface)[13]. The photons have energies of up to hundreds of megaelectronvolts (MeV).

High-energy electrons, and γ-rays that have energies larger than about 10 MeV, can knock out neutrons from atmospheric nitrogen-14 and oxygen-16 nuclei - by electrodisintegration in the case of electrons and photonuclear reactions in the case of γ-rays[11,12]. Although the ability of thunderstorms to produce neutrons through photonuclear reactions has been demonstrated using computer simulations[11,13], direct experimental evidence has been absent.

Rather than focusing on the neutrons, Enoto and colleagues considered the other products of the photonuclear reactions involving nitrogen-14 and oxygen-16: namely, unstable nitrogen-13 and oxygen-15 isotopes (Fig. 1). These isotopes decay after a few minutes into stable carbon-13 and nitrogen-15 nuclei through the emission of a neutrino and a positron - the antiparticle of the electron. Finally, the positron annihilates with an electron of an atmospheric molecule to produce a pair of γ-rays.

Figure 1 | Nuclear reactions triggered by a thunderstorm. Enoto et al.2 report conclusive evidence that thunderstorms can induce nuclear reactions in the atmosphere. For example, the authors find that a thunderstorm can generate a high-energy ?-ray that knocks a neutron out of a nitrogen-14 nucleus, creating an unstable nitrogen-13 isotope. The isotope decays into a neutrino, a positron (the antiparticle of the electron) and a stable carbon-13 nucleus. Finally, the positron annihilates with an electron of an atmospheric molecule to produce a pair of ?-rays, each of which has a characteristic energy (0.511 megaelectronvolts).

Because both positrons and electrons have masses of 0.511 MeV (expressed in energy units), each emitted γ-ray has an energy of 0.511 MeV. Therefore, to confirm the existence of these photonuclear reactions, the authors simply needed to identify a line at this energy in the wide energy spectrum of all γ-rays.

To this end, Enoto et al. carried out ground-based observations of γ-ray emission from low winter thunderclouds above the coast of the Sea of Japan. On 6 February 2017, they detected an intense γ-ray flash that lasted for less than 1 millisecond, which they associated with a lightning stroke. After the initial γ-ray flash, the authors observed a prolonged γ-ray line at an energy of 0.511 MeV that lasted for about a minute (see Fig. 4 in the paper2). This line is a conclusive indication of electron-positron annihilation, and represents unequivocal evidence that photonuclear reactions can be triggered by thunderstorms.

Enoto and colleagues' discovery is important because it unveils a previously unknown natural source of isotopes in the atmosphere, in addition to the irradiation of Earth by cosmic rays. These isotopes include nitrogen-15, carbon-13 and carbon-14, the last of which is widely used in the dating of archaeological artefacts and artworks. In fact, the contribution of thunderstorms to Earth's carbon-14 abundance could be comparable in some regions to that of cosmic irradiation[14]. Future studies should check whether thunderstorms produce other isotopes (such as those of hydrogen, helium and beryllium).

Thunderstorm-induced nuclear reactions could occur in the atmospheres of other planets, such as Jupiter and Venus, and might therefore contribute to the isotopic composition of these atmospheres. However, determining the magnitude of this contribution will require detailed observations of γ-rays and neutrons from thunderstorms on these planets. Another implication of Enoto and colleagues' discovery is that the neutrons are formed outside the plasma created by lightning. This suggests that these neutrons cannot provide information about the plasma, in contrast to expectations[15].

Leonid Babich is at the Russian Federal Nuclear Center - All-Russian Research Institute of Experimental Physics (RFNC-VNIIEF), Sarov, Nizhni Novgorod region, 607185 Russia.
e-mail: leonid.babich52@gmail.com

1. Wilson, C. T. R. Proc. Cambridge Phil. Soc. 22, 534-538 (1925).
2. Enoto, T. et al. Nature 551, 481-484 (2017).
3. Chadwick, J. Nature 129, 312 (1932).
4. Gurevich, A. V., Milikh, G. M. & Roussel-Dupre, R. Phys. Lett. A 165, 463-468 (1992).
5. Eddington, A. S. Nature 117, 25-32 (1926).
6. Shah, G. N., Razdan, H., Bhat, C. L. & Ali, G. M. Nature 313, 773-775 (1985).
7. Chilingarian, A. et al. Phys. Rev. D 82, 043009 (2010).
8. Gurevich, A. V. et al. Phys. Rev. Lett. 108, 125001 (2012).
9. Tsuchiya, H. et al. Phys. Rev. D 85, 092006 (2012).
10. Ishtiaq, P. M., Mufti, S., Darzi, M. A., Mir, T. A. & Shah, G. N. J. Geophys. Res. Atmos. 121, 692-703 (2016).
11. Babich, L. P., Bochkov, E. I., Kutsyk, I. M. & Zalyalov, A. N. JETP Lett. 97, 291-296 (2013).
12. Babich, L. P. JETP Lett. 84, 285-288 (2006).
13. Dwyer, J. R., Smith, D. M. & Cummer, S. A. Space Sci. Rev. 173, 133-196 (2012).
14. Babich, L. P. Geophys. Res. Lett. 44, http://dx.doi.org/10.1002/2017GL075131 (2017).
15. Fleisher, R. L., Plumer, J. A. & Crouch, K. J. Geophys. Res. 79, 5013-5017 (1974).

Sulfurous climate on cold fusion

Cold Fusion is an inappropriate term that has done a tremendous harm to the LENR cause. Indeed, it is not a question of the fusion of two atoms, but of the transmutation of atoms bombarded by particles with high energy (neutrons, electrons, protons, etc.). There were also charlatans with their magic powder that was supposed to produce heat. Some are now in the courts. All this maintained a sulphurous climate around the cold fusion. This term has mostly become a means of derision used by traditional scientists

Hot fusion has been promised by scientists for 50 years. They invested in research billions of francs. They managed to build prototypes called Tokamak, the EPFL TCV for example, where a plasma of several million degrees simulates the conditions of the sun and allows the hydrogen atoms, for example, to merge into Helium and provide considerable clean energy (no CO2 or radioactivity). But this plasma is very difficult to control because no material is able to contain it. To do this, it is necessary to use extremely powerful magnetic fields. The Cadarache test plant in France with its Tokamak ITER is not yet ready to provide the energy expected from it (*).

Our approach is strictly scientific. Our three scientists are professors of theoretical physics. One of them is well known at Northeastern University in Boston, Massachusetts, the other two have a long career behind them at Northeastern University in Boston and at the University of Perugia in Italy. They are the authors of LENR electroweak and electro-strong theory. With these scientists, theory precedes experimentation. They have chosen simple experiments in our MIC laboratory (Marly Innovation Center) to clearly demonstrate the existence of LENR. Some very detailed measures were made at the CSEM in Neuchâtel. Our experiments and measurements can be repeated simply in any laboratory, almost on a kitchen table.

In an article, we show on the one hand that transmutations take place on the electrodes during a simple electrolysis and, on the other hand, that the explosion of a Lithium ion attery has a nuclear component. (Coulomb explosion). The important thing here is not to know what transmutation is done during the electrolysis, but simply that it is one; that the presence of new elements after the electrolysis is explicable only by the transmutation of other elements present before the electrolysis. We believe that it has now been shown that transmutations at normal temperatures are possible, and even common in nature. Alberto Carpinteri, professor at the Politecnico of Turin has published a series of articles demonstrating that the compression of the rocks produces neutrons and, consequently, nuclear transmutations on the fracture surfaces. He further demonstrated that earthquakes could be predicted one or more weeks before their onset, using neutron measurements 100 m or more deep.

But official science is slow to recognize what we think is obvious. It is true that we have all the difficulties to publish our article. Obviously he is disturbing. Almost all the scientific journals answered negatively on the pretext that it was not in the editorial line of their newspaper, without even taking the advice of a scientific referent. It was necessary to insist again and again until one of them, EFM (Engineering Fracture Mechanics), ended up sending the draft article to two referring scientists. Both referees responded very positively. Their remarks even allowed us to further improve the article from an educational point of view. The article is now published.

Simultaneously, we received confirmation that one of our scientists was invited to present our work for 20 minutes at the 18th Lomonosov Conference on Elementary Particle Physics, from 24-30 August 2017 at Moscow State University, then at the Annual Conference of Italian Physicists, in September 2017 in Trento. Something is changing.

We therefore hope that our universities and academic institutions in Switzerland will eventually become interested in these phenomena and will support us in our research that is oriented in several directions. (1) Describe the nuclear reactions responsible for the explosion of Lithium Ion batteries and avoid these dangerous explosions (public safety problem). (2) To control certain transmutations, to find the most favorable ones and to generate abundant and clean energy (without CO2 or radioactivity). (3) Develop a reactor that, through controlled and harmless transmutations, produces the electricity needed to continuously recharge batteries, for example the batteries of an electric car. (4) Miniaturize this battery and its charging system to equip connected objects. (5) Find and master transmutations that would be able to turn highly radioactive waste into non-radioactive waste.

Serious scientists working in this field become very rare because no university produces them. One of the first actions would therefore be to create postdoctoral programs in this field. Our three scientists are among the world's top ten in this field. We are looking for financial support to carry out these projects.

(*) Editor's note: The JAM publishes here an opinion little consensual. Re-read in this regard the interview we gave Ambrogio Fasoli, in our previous edition (JAM 10, page 22). The director of the Swiss Plasma Center, also head of the Iter project for Switzerland, believes that the hot fusion keeps all its chances and that the Iter project should see the day in 2025 in Aix-en-Provence and that the delays are due for the essential to the mode of international cooperation. It takes longer but maximizes the chances of (political) success in the long run. Supported by the Confederation and the Swiss Nuclear Forum, he recalled that the Tokamak of Lausanne already manufactures at the moment plasma, whose temperature exceeds that of the sun. According to Professor Fasoli, these developments will also benefit Swiss companies.

The opinions expressed in this section are the sole responsibility of the author: Georges de Montmollin.

Researchers under pressure

THE THROUGH THE SCIENTIFIC PUBLICATION (1/5) Specialized journals and their publishing system heavily influence research. Researchers are encouraged to produce, not to discover.
CATHERINE FRAMMERY Facebook @cframmery

Peter Higgs, the physicist father of the famous boson that bears his name, could no longer do his research today. Because he did not find fast enough and did not publish enough, he himself said after obtaining his Nobel Prize in 2013. The University of Edinburgh was on the brink of losing service when it was first nominated in 1980, which ultimately ensured it was tolerated to its price.

Publish? If science progresses through research, it has depended since the seventeenth century on scientific journals for dissemination. It is in Nature, Science, Cell, The Lancet and others that are announced and discovered, commented on, taken up and completed. And it is thus by publishing that a researcher will be able to prove his qualities to his community. But the machine is packed. It is as if universities and funding bodies have delegated to scientific journals the task of sorting out projects, ideas and researchers. Scientific articles have become the criterion that makes you believe in yourself and give you money, a job, a promotion. Or not. A grueling race for publication has begun all over the world. More than 2.5 million articles are written each year. "The hamster in his wheel", blows a young biologist.

Publish or perish
Many students or confirmed researchers have answered our call for testimonials. Sign of their concern, most asked us not to mention their name. It is not good to criticize a system in which one dreams of entering. "We do research not to understand the world, but to publish," regrets this post-doc. It is advisable to publish at the master's degree, to increase its chances of doing a doctorate. "For my supervisor, it's a good way to know if a student will be productive," says a PhD student in biology. The pressure increases in doctorate. "No article, no doctorate. Because the more publications a university has, the higher it gets in the Shanghai ranking » (a world ranking of the most prestigious universities established by Shanghai Jiao Tong University), according to another doctoral student. The theses have become lists of publications.

The race continues then. Post-docs, with precarious contracts, are also obsessed with their articles to publish to continue to have subsidies, find a salary, a job. "One of my former colleagues escaped thanks to a private foundation, another finished the writing of his thesis and his publications while being unemployed", denounces this doctoral student. Finally, even appointed, academics remain under pressure from publication to be confirmed, to show their employer that they can bring money, or simply to carry out their projects. "We increase the teaching load of teachers who publish too little, or they add administrative tasks," says an American mathematician.

"Impact factor", my love
But there is publish and publish. More than 25,000 scientific journals exist today, based on peer review, of which 12,000 have an "impact factor" and are ranked in the Thomson Reuters ranking. This impact is based on the average number of citations of an article of the journal during the two previous years. It is currently 47 for The Lancet, 40 for Nature and 37 for Science, and can be less than 1 for small hyperspecialized journals. All the difficulty for a researcher will be to aim for the "good" level of review for his article, by evaluating the risks of rejection and the possible gains in terms of recognition.

Bibliometry took power, this comfortable habit of identifying researchers to numbers, such as the ubiquitous and controversial "h-index," which takes into account their number of articles and the number of citations of those articles in other studies. In fact, this index disadvantage authors of few articles, even if they are important. That of Peter Higgs, always him, is thus 11, while at the American Academy of Sciences, the scores are often greater than 40.

The damage of this "publish or perish" is first human, we guessed it. Posting is a huge job. The writing takes months of adjustments, of struggle with all the authors. And, even interested, a review almost always asks for details, new experiences, and therefore new months of work and stress. In the meantime scholarships have come to an end, other teams have advanced. Researchers are reluctant to have children. "A colleague's supervisor says that she must take advantage of her maternity leave to work on her next articles," wrote a Genevan biologist. Between exhaustion and renunciation, the human waste is there. But the damage is also scientific. Because this crazy race for publication leads to many articles of less quality, that fewer and fewer researchers have time to read. We dabble results to get three items where one would have been enough. We exaggerate the importance of a study to create a "wow effect" that will seduce. Pups are cited to attract benevolence. We are too aggressive to "go up" in the search engines. We prefer fashionable subjects.

Results hacked Above all, we put in scene of supposed mechanisms of explanation by doing the storytelling at the expense of the confirmation of the principal observations, base of the science. "Flemming (who accidentally discovered penicillin, Ed) could not publish today," indignant Lawrence Itegendran, founder of the platform Science Matters, which argues for a science less theatrical. No wonder that the poor reproducibility of science has become a major concern in recent years. Not to mention rising fraud. We hack results so that they go in the desired direction, knowing that few teams will want to take the time to check results already published. Or you pay your ticket to be added as an author in unscrupulous journals - as revealed at the Peer Review Congress in Chicago last week. All these excesses make the scientific community react. In 2013, the Declaration on Research Evaluation (DORA), which advocates dropping impact factors to judge a researcher, was signed in San Francisco. "It is a very unscientific, naive and inadequate method to take a review as a measure," says Matthias Egger, president of the Swiss National Science Foundation, determined to implement the procedures from DORA. Change also involves digital disruption, with the advent of open access and open science. And it is accelerated by the greed of scientific publishers, whose juicy profits provoke a sling in Germany and Finland. The purse of the States, new armed arm of the laboratories. This is our next episode.

Tomorrow: The business of scientific publishing

First author, last author, the big trouble

Who really produced the scientific article you read? Investigation

Articles are almost always the result of teamwork. Traditionally, the first author has generated the majority of the data and the last name cited is that of the PI, the Principal Investigator, who imagined, directed the work and ensured the follow-up of the latter. Between the two ... it's more vague. There are those who contributed by donating equipment, premises or technicians, but also those who are quoted prodeo for later obtaining a "lift return" and those who take advantage of their position of strength to enrich their list of publications and quotes - any article allows you to promote your "h-index" (see above) and therefore your career, even if you are in the middle of the authors.

Rules anyway
The dirty shots abound. History has marked Raphaël Grolimund, EPFL librarian in Lausanne, who advises the students: it is thanks to him that a grieving doctoral student discovered, too late, that his thesis director should never have been write in his place first name in his article. Because in Europe "there are rules, it's not just a balance of power". To sign an article, you must have participated in the research, the writing of the article, and agreed to the final text. Universities have mediators or ethics committees to resolve disputes. Bosses of labs cited as authors in hundreds of publications that they have not even read should no longer exist. The order can be alphabetical when it comes to big collaborations: the article highlighting the gravitational waves counts more than 1000 authors! Many researchers believe that the criteria used to classify authors' names should be systematically clarified. In China, India, an author's place in an article is directly related to funding: the article does not count if you are not in 1st or 2nd position. An additional violence of publish or perish

Small arrangements in the labs

THE THROUGH THE SCIENTIFIC PUBLICATION (4/5) In order to reach discoveries, researchers sometimes resort to dubious trickery: image fakes and statistical hacking abound. Fourth part of our series
DAVID LAROUSSERIE (LE MONDE)

Copy and paste images, statistical hacking, exaggeration of results, lack of knowledge of the methods used, slow or even refusal to correct errors the backyards of laboratories are not always brilliant. In June 2016, in the journal mBio, a screening of more than 20,000 articles from 40 scientific journals found nearly 4% of issues with images in the articles supporting the demonstrations. The rate exceeds 12% for a sample newspaper. "Errors" range from simply duplicating parts of images, to fraudulent retouching through repositioning or inversion of parts. These images show, essentially, which proteins are expressed or not in tissues.

The Retraction Watch database, a website launched in 2010 to track the news of withdrawals or corrections of articles, identifies more problematic cases for "manipulation" of images than for "text plagiarism" (plagiarism of images also exist!): 294 plagiarism of articles for 422 duplications, 305 manipulations and 134 falsifications of images. Another site, PubPeer, launched in 2012 to host anonymous discussions on previously published articles, quickly turned into a forum for tracking manipulated images. Which led to many corrections and retractions.

Retouched images One of the dramas is "that reviewers do not look at the images," notes Elisabeth Bik, a microbiologist with the microbial genomics company uBiome in California and co-author of the mBio study. She spotted the mistakes before other colleagues validated them. It also points to another problem: the absence of reactions from the authors or the newspapers that publish the disputed articles. She estimates that she has reported more than 800 cases that led to about 30 retractions, "but in the vast majority of cases, I have not had any answers". The specialist, to explain these practices more or less questionable, evokes "the error, the lack of time to make the experiments of control, the precipitation to publish or the desire to hide things ..." It has also fallen on recidivists having more than twenty retouched images, evidence of more serious dysfunctions. In a new article to appear, she highlighted correlations. The pressure to publish increases the risk of bad practice, while greater social control, ie the existence of rules or sanctions, limits it. To solve these problems, the researcher is involved in the development of software for automatic detection of image editing, which are beginning to be developed publishers.

The art of "p-hacking"
Researchers also know how to deal with statistics, the tool that is used to analyze their results and that allows especially to claim a discovery (the lack of discovery is rarely published). On September 1st, more than seventy researchers called in Nature Human Behavior to "redefine statistical significance". For them, "the statistical standards for claiming a discovery are simply too low in many areas of science". And they call to raise these standards.

Starting with the best known of them, the value p. The "standard" means that a statistical test measuring the difference between two hypotheses and giving a p value of less than 5% is significant and therefore worthy of publication. First problem, for years, researchers have warned that some people ignore the very definition of this value p. Many believe that this parameter refers to the probability that an experimental result is a false positive. But that's not really the case.

David Colquhoun of the University College in London explained in 2014 in an article of the Royal Society, with the example of a test of detection of a disease. A p-value of 5% means that if someone is not sick, then the test will find that it has a 5% chance of being (false positive). But that does not say it's the probability of being sick. By taking a prevalence rate of 90% for example for this disease one can then calculate the real rate of false positive as being 36%! The value p alone can therefore lead to misinterpretations. Nevertheless, the lower a threshold is set, the lower the false positive rate. Same if you increase the size of the sample.

But while genetics or physics have set much more drastic thresholds for p (ten to one hundred millionths), disciplines such as biomedical research, psychology, economics ... remain attached to this 0.05. In March 2016 a study by John Ioannidis in JAMA noted the presence of p-value in the summary of one-third of the top 151 medical journals and in nearly 40% of clinical trials. Small oddity, already noted by others: reported p-values ??have a strong tendency to focus to 0.05, the famous threshold from which the results are considered significant. It is undoubtedly that researchers are masters in the art of "p-hacking", that is to say the art of finding the right method to fall below the fateful threshold.

"Overexploitation" of the data
"Some people are over-exploiting the data and trying until it works," says Bertrand Thirion, neuroscience specialist at Inria, the French National Institute for Research in Computer Science and Automation. "It's not deliberate cheating, but as the researchers have put a lot of effort into doing the experiments, they want to find something and" vibrate "the methods. Chris Chambers, in his book The Seven Deadly Sins of Psychology (Princeton University Press, untranslated) details with regret these bad practices. "The effects of p-hacking are clear, filling the scientific literature with assumptions made after the experiment, false discoveries, and research stalemates," he writes.

To improve reliability, the authors of the call for Nature human behavior first recommend lowering the threshold to 0.005 and also mention the existence of other criteria or statistical methods. This problem of the value p is strongly linked to a plague of research, "the crisis of the reproducibility" ... to discover in the next part of our series.

The "spin", or how to twist reality

When one has tried the forceps to arrive at a result (copy-paste of images, exploitation of the statistical methods ...), but that the nature still persists to prevent a discovery, it remains the solution of the sleight of hand: present the conclusions in a more beautiful light than reality. This spin, as the Anglo-Saxons say, can be done during the passage from the article to the press release, but also from the scientific article to its summary by the authors (on which many readers stop), as have highlighted many works.

So on September 11 in Pios Bioiogy a team from the University of Sydney analyzed 35 studies on spin and confirmed that the practice is widespread. In the articles reporting clinical trials, more than half, 57%, present distortions of reality. In 2014, another study found that 40% of press releases contain exaggerations and that in these cases, 5 8% of newspaper articles mentioning these works exaggerate too ...

The tricks of the game involve over-interpreting the conclusions, or exaggerating the scope of a statistical test. Or to put forward a secondary result to suggest a benefit of a treatment. Or to attribute, without proof, a cause to an effect.

A science impossible to reproduce

THE THROUGH THE SCIENTIFIC PUBLICATION (5/5) The ability to redo and confirm an experiment is at the heart of the scientific process. Yet in a lot of cases it does not work. Last episode of our series
DAVID LAROUSSERIE (LE MONDE)

There is danger in delay. The "immune system of science" is failing, as psychologist Chris Chambers notes in his book, The 7 Deadly Sins of Psychology (The Seven Deadly Sins of Psychology, Princeton University Press, Untranslated). The learned name of this protection is "reproducibility", that is to say the possibility of redoing and confirming an experiment.

"Reproducibility and replication are the cornerstones of science. Without them, we have no way of knowing which discoveries are true and which are caused by chance, error or fraud, "says Chris Chambers of Cardiff University. We do not fly in an airplane that has not been rigorously tested over and over again. The same goes for all branches of science."

Most of the time, it works. Thus in 2014, faced with the impossibility of several laboratories to easily repeat a protocol claiming to obtain pluripotent stem cells, the Japanese team that claimed the discovery in Nature is forced to admit that it has defrauded. The culprit, Haruko Obokata, has resigned and one of his co-authors, Yoshiki Sasai, yet innocent, will commit suicide. Ditto for a genome editing technique, which promised to do better than the technique Crispr-case9, very popular. The article published by Nature Biotechnology in 2016 was withdrawn in August, after the failure of several teams to reproduce the result.

System failure
Contrary to these "successes", in 2005, Stanford University's John Ioannidis shook the community with an article in Plos Medicine suggesting that "most scientific results are false" because they are impossible to reproduce. Many replication experiments have since been conducted, showing the extent of system failure. In 2012, a team from the biotechnology company Amgen explains that in oncology it has found the published results in only 6 out of 53 cases. In 2015, the first initiative of the Center for Open Science in the United States, tries to reproduce 100 experiences of psychology and succeeds in 39 cases. Two years later, an identical replication program in cancerology publishes its first results. Of 7 studies, 4 have been reproduced, 1 no and 2 remain impossible to interpret. There are still 29 studies to check in this project. Often, it is the misinterpretation of statistical tests that makes them difficult to replicate.

Call for more rigor
Further proof of the malaise, in May 2016, Nature published a devastating survey: 70% of the 1,576 respondents say they have failed to reproduce a result and even 50% to redo their own experience ... The time is so bad that in January 2017 ten authors signed a "manifesto for reproducible science" in Nature Human Behavior calling for more rigor in methods, reporting and evaluation of research.

The "crisis", as it was baptized, does not only affect psychology or oncology. In functional brain MRI imaging, several studies have shown that pixel activations deemed significant are actually false positives. One of these studies shows that there are so many ways to analyze images that it is possible to "turn on" 90% of the pixels of a scanner by applying one or the other. "In my team, we test the methods on several datasets to avoid these pitfalls," says Bertrand Thirion, Inria, the French National Institute for Research in Computer Science and Automation.

In chemistry, Raphaël Lévy of the University of Liverpool quotes in the journal Médecine / Sciences of 18 September three cases of results probably wrong in his discipline, which have not been corrected for several years, despite its challenges and those of others researchers. "The system does not encourage replication of results. It takes a little crazy to engage in these criticisms, even if it does not hurt my career, "says the researcher. Scientific journals are not always quick to correct published errors, which undermine their reputation.

"The public has the right to build trust in science over reality and not fiction. Even with these flaws, science is undoubtedly the best way to discover the truth about the world and make rational decisions. But that does not mean that it can not or should not be improved. We need to find practical solutions to these problems, "said Chris Chambers.

Access to data
The list which he proposes in his book, or which is taken up by the manifesto which he co-signed, is teeming with ideas. Like raising the requirements for statistical rigor. Or promote transparency in procedures, by giving access to raw data, images, figures, methods used .. "There are still brakes on this opening. For some, data is power. For others, it is the fear that we find defects in their work, regrets Bertrand Thirion. But it's good to find mistakes so that you can correct them! " In 2016, Chris Chambers and others launched a charter for article reviewers who commit to evaluating manuscripts only if the authors submit their data.

Another solution consists in pre-registration of experience, as practiced for several years for clinical trials. Researchers should detail their protocol and the methods they will use for their experiment, to avoid the temptation to adapt the method or tests to the observations. Sites such as the Open Science Framework, launched by the Center for Open Science, now make it easy to fill out such recommendations. Another idea, defended by provocation by John Ioannidis in 2014: to reverse the usual "incentives" for researchers to promote replication studies, data sharing, rather than excessive publication.

"We have a heavy public responsibility to ensure that the next generation of scientists does not suffer from the problems of my generation. Only when science is as open and solid as possible can it bring maximum benefit to humanity, "concludes Chris Chambers.

In ecology, we anticipate the problem

RESEARCH Fourteen European laboratories tried to replicate the same experiment on legumes, but did not succeed. But they have developed a parade to variability due to the specificities of each laboratory.

In ecology, it is not yet the crisis of reproducibility, but we are preparing for it. An article, available since August on BioRxiv, reports the collaboration of 14 laboratories in Europe to test the robustness of their field in the face of this difficulty to validate certain published results. The idea was to test whether the same experiment, namely the effect of a legume on the growth of a grass planted jointly, could be strictly reproduced in different laboratories under the same conditions.

"There was some evidence that the reproducibility is lower than expected, says Alexandru Milcu, Ecotron Montpellier and the Center for Functional and Evolutionary Ecology (CNRS), at the origin of this collaboration. There are specific laboratory conditions that escape us, such as the nature of the microenvironment, the role of experimenters ... " Finally, this variability has been observed. But the team also found a way to multiply the experiments with plants with different genotypes. This is counterintuitive, but this added and controlled variability "drowns" the specifics of the place and increases the reproducibility of the experiments. "We will have to repeat this kind of study. What at first was a question of curiosity has become fascinating! The researcher notes.

Climat sulfureux sur la fusion froide

Fusion froide est un terme inadéquat qui a fait un tort énorme à la cause LENR. En effet, il ne s'agit pas de la fusion de deux atomes, mais de la transmutation d'atomes bombardés par des particules à haute énergie (neutrons, électrons, protons, etc.). Il y a aussi eu des charlatans avec leur poudre magique qui devait produire de la chaleur. Certains sont maintenant devant les tribunaux. Tout ceci a entretenu un climat sulfureux autour de la fusion froide. Ce terme est surtout devenu un moyen de dérision utilisé par les scientifiques traditionnels.

La fusion chaude nous est promise par les scientifiques depuis 50 ans. Ils ont investi en recherche des milliards de francs. Ils ont réussi à construire des prototypes nommés Tokamak, le TCV de l'EPFL par exemple, où un plasma de plusieurs millions de degrés simule les conditions du soleil et permet effectivement aux atomes d'Hydrogène, par exemple, de fusionner en Hélium et de fournir une énergie propre considérable (pas de CO2, ni de radioactivité). Mais ce plasma est très difficilement contrôlable, car aucun matériel n'est capable de le contenir. Pour ce faire, il faut utiliser des champs magnétiques extrêmement puissants. La centrale test de Cadarache en France avec son Tokamak ITER n'est pas encore prête à fournir l'énergie qu'on attend d'elle (*).

Notre démarche est strictement scientifique. Nos trois scientifiques sont des professeurs de physique théorique. L'un d'eux est bien connu à la Northeastern University à Boston (dans le Massachusetts), les deux autres ont une longue carrière derrière eux à la Northeastern University à Boston et à l'Université de Perugia en Italie. Ils sont les auteurs de la théorie LENR electroweak et electro-strong. Avec ces scientifiques, la théorie précède l'expérimentation. Ils ont choisi des expériences simples dans notre laboratoire du MIC (Marly Innovation Center) pour démontrer clairement l'existence de LENR. Certaines mesures très pointues ont été faites au CSEM à Neuchâtel. Nos expériences et nos mesures peuvent être répétées simplement dans n'importe quel laboratoire, presque sur une table de cuisine.

Dans un article, nous montrons d'une part que des transmutations ont lieu sur les électrodes lors d'une simple électrolyse et, d'autre part, que l'explosion d'une atterie au Lithium ion a une composante nucléaire (explosion de Coulomb). L'important ici n'est pas de savoir quelle transmutation se fait lors de l'électrolyse, mais simplement qu'il s'en fait une; que la présence de nouveaux éléments après l'électrolyse ne soit explicable que par la transmutation d'autres éléments présents avant l'électrolyse. Nous estimons qu'il est maintenant démontré que des transmutations à températures normales sont possibles, voire même courantes dans la nature. Alberto Carpinteri, professeur au Politecnico de Turin a publié une série d'articles démontrant que la compression des roches produit des neutrons et, par suite, des transmutations nucléaires sur les surfaces de fracture. Il a, de plus, démontré, que les tremblements de terre pouvaient être prévus une ou plusieurs semaines avant leur déclenchement, grâce à des mesures de neutrons, à 100 m ou plus de profondeur.

Mais la science officielle tarde à reconnaître ce qui nous semble être une évidence. Il est vrai que nous avons toutes les difficultés pour faire paraître notre article. A l'évidence il dérange. Presque tous les journaux scientifiques ont répondu négativement en prétextant que ce n'était pas dans la ligne éditoriale de leur journal, sans même prendre l'avis d'un référent scientifique. Il a fallu insister encore et encore jusqu'à ce que l'un d'eux, EFM (Engineering Fracture Mechanics), finisse par envoyer le projet d'article à deux scientifiques référents. Les deux référents ont répondu très positivement. Leurs remarques nous ont même permis d'améliorer encore l'article du point de vue pédagogique. L'article est maintenant publié.

Simultanément, nous recevions la confirmation que l'un de nos scientifiques était invité à présenter nos travaux pendant 20 minutes à la 18e conférence Lomonosov sur la physique des particules élémentaires, du 24-30 août 2017 à l'Université d'Etat à Moscou, puis à la Conférence annuelle des physiciens italiens, en septembre 2017 à Trente. Quelque chose est en train d'évoluer...

Nous avons donc l'espoir que nos hautes écoles et institutions académiques suisses finiront par s'intéresser à ces phénomènes et nous soutiendront dans nos recherches qui s'orientent dans plusieurs directions. (1) Décrire les réactions nucléaires responsables de l'explosion des batteries au Lithium ion et permettre d'éviter ces explosions dangereuses (problème de sécurité publique). (2) Maîtriser certaines transmutations, trouver les plus favorables et générer une énergie abondante et propre (sans CO2 ni radioactivité). (3) Développer ainsi un réacteur qui, par des transmutations maîtrisées et inoffensives, produise l'électricité nécessaire pour recharger continuellement des batteries, par exemple les batteries d'une voiture électrique. (4) Miniaturiser cette batterie et son système de recharge pour en équiper les objets connectés. (5) Trouver et maîtriser les transmutations qui seraient capables de transformer les déchets hautement radioactifs en déchets ordinaires non radioactifs.

Les scientifiques sérieux qui travaillent dans ce domaine deviennent très rares, car aucune université n'en produit. L'une des premières actions serait donc de créer des filières postdoctorales dans ce domaine. Nos trois scientifiques font partie du top ten mondial dans ce domaine. Nous cherchons des soutiens financiers pour réaliser ces projets.

(*) N.d.l.r.: Le JAM publie ici une opinion peu consensuelle. Relire à ce propos l'interview que nous a accordée Ambrogio Fasoli, dans notre édition précédente (JAM 10, page 22). Le directeur du Swiss Plasma Center, également patron du projet Iter pour la Suisse, estime que la fusion chaude conserve toutes ses chances et que le projet Iter devrait voir le jour en 2025 à Aix-en-Provence et que les retards sont dus pour l'essentiel au mode de coopération internationale. Cela prend plus de temps mais permet d'optimiser les chances de réussite (politique) sur le long terme. Soutenu par la Confédération et le Forum nucléaire suisse, il rappelle que le Tokamak de Lausanne fabrique déjà à l'heure actuelle du plasma, dont la température dépasse celle du soleil. Selon le professeur Fasoli, ces développements profiteront également aux entreprises suisses.

Les opinions exprimées dans cette rubrique n'engagent que l'auteur: Georges de Montmollin

Petits arrangements dans les labos

LES TRAVERS DE LA PUBLICATION SCIENTIFIQUE (4/5) Pour aboutir à des découvertes, les chercheurs ont parfois recours à des tours de passe-passe douteux: trucages d'images et bidouillages statistiques abondent. Quatrième volet de notre série
DAVID LAROUSSERIE (LE MONDE)

Copier-coller d'images, bidouillage statistique, exagération des résultats, méconnaissance des méthodes utilisées, lenteur voire refus à corriger des erreurs..., les arrière-cours des laboratoires ne sont pas toujours reluisantes. En juin 2016, dans le journal mBio, un criblage de plus de 20 000 articles tirés de 40 journaux scientifiques a repéré près de 4% de problèmes avec des images présentes dans les articles à l'appui des démonstrations. Le taux dépassant 12% pour un journal de l'échantillon. Les «erreurs» vont de la simple duplication de parties d'images, à la retouche frauduleuse en passant par le repositionnement ou l'inversion de certaines parties. Ces images montrent, pour l'essentiel, quelles protéines sont exprimées ou non dans des tissus.

La base de données de Retraction Watch, un site lancé en 2010 pour suivre l'actualité des retraits ou corrections d'articles, recense plus de cas problématiques pour «manipulation» d'images que pour «plagiat de texte» (le plagiat d'images existant également!): 294 plagiats d'articles pour 422 duplications, 305 manipulations et 134 falsifications d'images. Un autre site, PubPeer, lancé en 2012 pour accueillir des discussions anonymes sur des articles déjà publiés, s'est vite transformé en forum de la traque des images manipulées. Ce qui a conduit à bien des corrections et rétractions.

Images retouchées L'un des drames est «que les reviewers ne regardent pas les images» constate Elisabeth Bik, microbiologiste de la société de génomique microbienne uBiome en Californie et coauteur de l'étude de mBio. Elle a repéré à l'oeil les erreurs, avant que d'autres collègues ne les valident. Elle pointe aussi un autre problème: l'absence de réactions des auteurs ou des journaux qui publient les articles litigieux. Elle estime avoir signalé plus de 800 cas qui ont conduit à une trentaine de rétractions, «mais dans la grande mqjorité des cas, je n'ai pas eu de réponses». La spécialiste, pour expliquer ces pratiques plus ou moins discutables, évoque «l'erreur, le manque de temps pour faire les expériences de contrôle, la précipitation à publier ou l'envie de cacher des choses...» Elle est aussi tombée sur des récidivistes ayant plus d'une vingtaine d'images retouchées, preuve de dysfonctionnements plus graves. Dans un nouvel article à paraître, elle a mis en avant des corrélations. La pression à publier augmente le risque de mauvaises pratiques, tandis qu'un contrôle social plus important, c'est-à-dire l'existence de règles ou de sanctions, le limite. Pour résorber ces problèmes, la chercheuse est impliquée dans la mise au point de logiciels de détection automatique de retouches d'images, dont commencent à se doter les éditeurs.

L'art du «p-hacking»
Les chercheurs savent aussi s'arranger avec les statistiques, l'outil qui leur sert pour analyser leurs résultats et qui permet surtout de clamer une découverte (l'absence de découverte faisant rarement l'objet de publication). Le 1er septembre, plus de soixante-dix chercheurs ont appelé dans Nature Human Behaviour à «redéfinir la significativité statistique». Pour eux, «les standards statistiques pour revendiquer une découverte sont tout simplement trop bas dans beaucoup de domaines de la science». Et ils appellent à relever ces standards.

A commencer par le plus connu d'entre eux, la valeur p. Le «standard» veut qu'un test statistique mesurant la différence entre deux hypothèses et donnant une valeur p inférieure à 5% soit significatif et donc digne d'être publié. Premier problème, depuis des années, des chercheurs ont alerté sur le fait que certains ignorent la définition même de cette valeur p. Beaucoup croient ainsi que ce paramètre désigne la probabilité qu'un résultat expérimental soit un faux positif. Mais ce n'est pas vraiment le cas.

David Colquhoun de l'University Collège à Londres l'a expliqué en 2014 dans un article de la Royal Society, avec l'exemple d'un test de détection d'une maladie. Une valeur p de 5% signifie que si quelqu'un n'est pas malade, alors le test trouvera qu'il a 5% de chance de l'être (faux positif). Mais cela ne dit pas qu'elle est la probabilité d'être malade. En prenant un taux de prévalence de 90% par exemple pour cette maladie on peut alors calculer le taux réel de faux positif comme étant 36%! La valeur p seule peut donc induire de fausses interprétations. Néanmoins, plus on fixe un seuil bas, plus ce taux de faux positif baissera. Idem si on augmente la taille de l'échantillon.

Mais alors que la génétique ou la physique ont fixé des seuils autrement plus drastiques pour p (dix à cent millionièmes), des disciplines comme la recherche biomédicale, la psychologie, l'économie... restent accrochées à ce 0,05. En mars 2016 une étude de John Ioannidis dans JAMA notait la présence de valeur p dans le résumé d'un tiers des 151 revues médicales les plus importantes et dans près dé 40% des essais cliniques. Petite bizarrerie, déjà constatée par d'autres: les valeurs p rapportées ont une forte tendance à se concentrer vers 0,05, le fameux seuil à partir duquel les résultats sont considérés significatifs. C'est sans doute que les chercheurs sont passés maître dans l'art du «p-hacking», c'est-à-dire l'art de trouver la bonne méthode afin de tomber sous le seuil fatidique.

«Surexploitation» des données
«Certains surexploitent les données et essaient jusqu'à ce que ça marche», explique Bertrand Thirion, spécialiste en neurosciences à l'Inria, l'Institut national français de recherche en informatique et en automatique. «Ce n'est pas de la triche délibérée mais comme les chercheurs ont fait beaucoup d'efforts pour faire les expériences, ils veulent trouver quelque chose et font «vibrer», les méthodes». Chris Chambers, dans son livre Les sept péchés mortels de la psychologie (Princeton University Press, non traduit ) détaille avec regret ces mauvaises pratiques. «Les effets du p-hacking sont clairs, remplissant la littérature scientifique avec des hypothèses faites après l'expérience, de fausses découvertes, et des impasses de recherche», écrit-il.

Pour améliorer la fiabilité, les auteurs de l'appel de Nature human behaviour recommandent dans un premier temps de baisser le seuil à 0,005 et évoquent aussi l'existence d'autres critères ou méthodes statistiques. Ce problème de la valeur p est fortement lié à une plaie de la recherche, «la crise de la reproductibilité»... à découvrir dans le prochain volet de notre série.

Le «spin», ou comment tordre La réalité

Quand on a essayé le forceps pour arriver à un résultat (copier-coller d'images, exploitation des méthodes statistiques...), mais que la nature persiste encore à empêcher une découverte, il reste la solution du tour de passe-passe: présenter les conclusions sous un jour plus beau que la réalité. Ce spin, comme disent les Anglo-Saxons, peut se faire lors du passage de l'article au communiqué de presse, mais aussi de l'article scientifique à son résumé par les auteurs (sur lequel beaucoup de lecteurs s'arrêtent), comme l'ont mis en évidence de nombreux travaux.

Ainsi le 11 septembre dans Pios Bioiogy une équipe de l'Université de Sydney a analysé 35 études sur le spin et confirmé que la pratique est répandue. Dans les articles rapportant des essais cliniques, plus de la moitié, 57%, présente des gauchissements de la réalité. En 2014, une autre étude constatait que 4 0 % des communiqués de presse contiennent des exagérations et que dans ces cas-là, 5 8% des articles de journaux mentionnant ces travaux exagèrent aussi...

Les tours de passe-passe consistent à surinterpréter les conclusions, ou à exagérer la portée d'un test statistique. Ou bien à mettre en avant un résultat secondaire pour suggérer un bénéfice d'un traitement. Ou encore à attribuer, sans preuve, une cause à un effet.

Une science impossible à reproduire

LES TRAVERS DE LA PUBLICATION SCIENTIFIQUE (5/5) La possibilité de refaire et de confirmer une expérience est au coeur de la démarche scientifique. Pourtant dans un grand nombre de cas, cela ne fonctionne pas. Dernier épisode de notre série
DAVID LAROUSSERIE (LE MONDE)

Il y a péril en la demeure. Le «système immunitaire de la science» connaît des ratés, comme le constate le psychologue Chris Chambers dans son livre, The 7 Deadly Sins ofPsychology (les 7 péchés mortels de la psychologie, Princeton University Press, non traduit). Le nom savant de cette protection est «reproductibilité», c'est-à-dire la possibilité de refaire et de confirmer une expérience.

«La reproductibilité et la réplication sont les pierres angulaires de la science. Sans elles, nous n'avons aucun moyen de savoir quelles découvertes sont vraies et lesquelles sont causées par le jeu du hasard, de l'erreur ou de la fraude, précise Chris Chambers, de l'Université de Cardiff. On ne vole pas dans un avion qui n'a pas été rigoureusement testé, encore et encore. I l en va de même pour toutes les branches de la science.»

La majorité du temps, cela fonctionne. Ainsi en 2014, devant l'impossibilité de plusieurs laboratoires à répéter aisément un protocole prétendant obtenir des cellules souches pluripotentes, l'équipe japonaise qui avait clamé la découverte dans Nature est contrainte d'avouer qu'elle a fraudé. La fautive, Haruko Obokata, a démissionné et l'un de ses coauteurs, Yoshiki Sasai, pourtant innocenté, se suicidera. Idem pour une technique d'édition du génome, qui promettait de faire mieux que la technique Crispr-cas9, très en vogue. L'article publié par Nature Biotechnology en 2016 a été retiré en août, après l'échec de plusieurs équipes à reproduire le résultat

Défaillance du système
A l'inverse de ces «réussites», en 2005, John Ioannidis, de l'Université de Stanford, ébranlait la communauté par un article dans Plos Medicine suggérant que «la plupart des résultats scientifiques sont faux», car impossibles à reproduire. De nombreuses expériences de réplication ont depuis été conduites, montrant l'ampleur de la défaillance du système. En 2012, une équipe de la société de biotechnologie Amgen explique qu'en oncologie elle n'a retrouvé les résultats publiés que dans 6 cas sur 53. En 2015, la première initiative du Centre pour la science ouverte aux Etats-Unis, tente de reproduire 100 expériences de psychologie et n'y parvient que dans 39 cas. Deux ans plus tard, un programme identique de réplication en cancérologie publie ses premiers résultats. Sur 7 études, 4 ont été reproduites, 1 non et 2 restent impossibles à interpréter. Reste encore 29 études à vérifier dans ce projet. Souvent, c'est la mauvaise interprétation de tests statistiques qui les rend difficiles à répliquer.

Appel à plus de rigueur
Preuve supplémentaire du malaise, en mai 2016, Nature publiait un sondage dévastateur: 70% des 1576 répondants déclarent avoir échoué à reproduire un résultat et même 50% à refaire leur propre expérience... L'heure est si grave qu'en janvier 2017 dix auteurs signent un «manifeste pour la science reproductible» dans Nature Human Behaviour appelant à plus de rigueur dans les méthodes, les comptes rendus et l'évaluation de la recherche.

La «crise», comme elle a été baptisée, ne touche pas seulement la psychologie ou l'oncologie. En imagerie cérébrale par IRM fonctionnelle, plusieurs études ont montré que des activations de pixels jugées significatives sont en réalité des faux positifs. L'une de ces études montre qu'il existe tellement de manières d'analyser les images qu'il est possible d'«allumer» 90% des pixels d'un scanner en appliquant l'une ou l'autre. «Dans mon équipe, nous testons les méthodes sur plusieurs jeux de données afin d'éviter ces pièges», explique Bertrand Thirion, de l'Inria, l'Institut national français de recherche en informatique et en automatique.

En chimie, Raphaël Lévy de l'Université de Liverpool cite dans la revue Médecine/Sciences du 18 septembre trois cas de résultats probablement faux dans sa discipline, qui n'ont pas été rectifiés depuis plusieurs années, malgré ses contestations et celles d'autres chercheurs. «Le système n'encourage pas à la réplication des résultats. Il faut être un peu fou pour s'engager dans ces critiques, même si ça ne nuit pas à ma carrière», témoigne le chercheur. Les revues scientifiques ne sont en effet pas toujours promptes à corriger les erreurs publiées, qui entament leur réputation.

«Le public a le droit de fonder sa confiance en la science sur la réalité et non sur la fiction. Même avec ces défauts, la science est sans aucun doute le meilleur moyen de découvrir la vérité sur le monde et de prendre des décisions rationnelles. Mais cela ne veut pas dire qu'elle ne peut pas ou ne devrait pas être améliorée. Nous devons trouver des solutions pratiques face à ces problèmes», estime Chris Chambers.

Accès aux données
La liste qu'il propose dans son livre ou qui est reprise par le manifeste qu'il a cosigné fourmille d'idées. Comme relever les exigences en matière de rigueur statistique. Ou favoriser la transparence dans les procédures, en donnant accès aux données brutes, images, chiffres, méthodes utilisées... «Il y a encore des freins face à cette ouverture. Pour certains, les données, c'est le pouvoir. Pour d'autres, c'est la peur qu'on trouve des défauts dans leur travail, regrette Bertrand Thirion. Mais justement, c'est bien de trouver des erreurs, pour pouvoir les corriger!» Chris Chambers et d'autres ont d'ailleurs lancé en 2016 une charte pour les relecteurs d'articles qui s'engagent à n'évaluer des manuscrits que si les auteurs transmettent leurs données.

Une autre solution consiste dans les pré-enregistrements d'expérience, comme pratiqué depuis plusieurs années pour les essais cliniques. Les chercheurs doivent détailler leur protocole et les méthodes qu'ils utiliseront pour leur expérience, afin d'éviter la tentation d'adapter la méthode ou les tests aux observations. Des sites comme l'Open Science Framework, lancé par le Centre pour la science ouverte, permettent désormais de remplir facilement ce genre de recommandations. Autre idée, défendue par provocation par John Ioannidis en 2014: renverser les «incitations» habituelles des chercheurs pour promouvoir les études de réplication, le partage des données, plutôt que la publication à outrance...

«Nous avons la lourde responsabilité publique de veiller à ce que la prochaine génération de scientifiques ne souffre pas des problèmes de ma génération. Ce n'est que lorsque la science est aussi ouverte et solide que possible qu'elle peut apporter le maximum d'avantages à l'humanité», conclut Chris Chambers.

En écologie, on anticipe le problème

RECHERCHE Quatorze laboratoires européens ont tenté de reproduire la même expérience sur des légumineuses, sans y parvenir. Mais ils ont mis au point une parade à la variabilité due aux spécificités de chaque laboratoire.

En écologie, ce n'est pas encore la crise de la reproductibilité, mais on s'y prépare. Un article, disponible depuis août sur BioRxiv, relate la collaboration de 14 laboratoires en Europe pour tester la robustesse de leur domaine face à cette difficulté à valider certains résultats publiés. L'idée était de tester si la même expérience, à savoir l'effet d'une légumineuse sur la croissance d'une graminée plantée conjointement, pouvait être strictement reproduite dans différents laboratoires dans les mêmes conditions.

«On avait quelques indices que la reproductibilité est inférieure à celle attendue, explique Alexandru Milcu, de l'Ecotron à Montpellier et du Centre d'écologie fonctionnelle et évolutive (CNRS), à l'origine de cette collaboration. Il y a des conditions de laboratoires spécifiques qui nous échappent, comme la nature du micro-environnement, le rôle des expérimentateurs...» Finalement, cette variabilité a bien été constatée. Mais l'équipe a aussi trouvé une parade consistant à multiplier les expériences avec des plantes aux génotypes différents. C'est contre-intuitif, mais cette variabilité ajoutée et contrôlée «noie» en quelque sorte les spécificités du lieu et augmente ainsi la reproductibilité des expériences. «Il faudra répéter ce genre d'étude. Ce qui au départ était une question de curiosité est devenu fascinant! » note le chercheur.»