Shulou(Shulou.com)11/24 Report--

DeepMind recently announced the latest progress of AlphaFold: 214 million protein structures of more than 1 million species have been predicted, covering almost all known proteins on the planet, refreshing our expectations for it.

The resurgence of the hot scene of the birth of AlphaFold 2 has once again sparked a heated debate on social media at home and abroad. However, as "insiders", researchers in the field of life science have mixed comments on the results announced by AlphaFold.

Previously, several scholars in the field of bioinformatics, such as Pan Yi, Zhou Yaoqi and Xu Dong, have said that there are some problems in the massive data updated in the AlphaFold protein structure database, such as structural instability and can not be applied to research.

Professor Tang Jian of the MILA Lab of the University of Montreal in Canada also told the Medical AI Nuggets that the protein predicted by AlphaFold has a limited impact on drug research and development.

Recently, Dr. Derek Lowe, a senior expert in the pharmaceutical industry in the United States, posted a complaint to the outside world, especially the AlphaFold which is popular in the media.

Dr. Derek Lowe is a graduate of Duke University and has worked at large pharmaceutical companies for more than 30 years in drug discovery programs for the treatment of schizophrenia, Alzheimer's disease, diabetes, osteoporosis and other diseases.

▲ Derek Lowe

Last week, Derek Lowe published an article on the website of the Royal Society of Chemistry (Royal Society of Chemistry). He pointed out clearly that AlphaFold will not bring about innovation in the field of drug research and development.

The following is Derek Lowe's article, Leifeng net did not change the original intention of the arrangement.

For a long time, protein structure prediction has been considered as one of the most difficult problems in computational biology.

But in the past year or two, AlphaFold has made remarkable progress in this area, predicting the tissue structure of most human proteins.

If it had been ten years ago, the result would have been like science fiction.

I don't want to deny AlphaFold's achievements, but some news reports misunderstand the significance of AlphaFold's achievements.

We haven't made a big leap forward in understanding why proteins fold in this way. Protein structures usually exist in the form of coils, rings or flakes, but why not continue to delve into them?

If we only do the research at the current level, we will not be able to find many secret answers.

We already have thousands of predictions of the structure of new proteins, most of which are correct. And, although there are some exceptions, they do seem to be mostly correct.

AlphaFold's algorithm will not work properly in the face of disordered protein regions. The whole computing technology of AlphaFold is based on finding analogies of known structures, and there is nothing AlphaFold can do without comparable structures.

Some disordered proteins can be arranged orderly under the influence of various proteins, but there are also some proteins that have never had an ordered structure under any conditions.

When the protein cannot form an ordered structure, it is beyond the computing power of AlphaFold.

▼ AlphaFold provides confidence in its structural predictions. Dark blue structures have higher confidence, while yellow and orange structures have lower confidence.

It should be emphasized that through AlphaFold, we get the prediction of protein structure, not the real protein structure.

AlphaFold is a very practical method for protein prediction, but the only way to determine its accuracy is to obtain the actual data of protein by X-ray, nuclear magnetic resonance or frozen electron microscope.

However, due to the flexibility of conformation, even the actual data can not fully represent its accuracy. This is where media reports exaggerate the impact of AlphaFold protein structure database on drug research and development.

In the presence of small molecular ligands, the protein structure will change and slide, sometimes subtle and sometimes violent, but AlphaFold can not predict these changes.

Algorithmic solutions to these problems may eventually be found, but so far, there are not enough protein structures that can bind to small molecular ligands. The quantity we need is very large.

There are about 20 different protein side chains to consider, but the number of small molecular structures is so large that it is almost infinite by comparison.

Also, it sounds harsh (though it's true): in the process of drug development, the understanding of the structure of proteins rarely affects the progress of research and development.

Because researchers usually run projects on the basis of tests using pure proteins or living cells. The test data represent whether the compound meets the requirements of the researchers and whether it performs better with the manufacture of the new compound.

The structure of the protein may give researchers some inspiration about what compounds to make next, but it may not help.

In the final analysis, real numbers from real biological systems are the most important. As the drug development program progresses, these figures cover pharmacokinetic, metabolic and toxicological tests, which cannot really be dealt with at the protein structure level.

The final waterfall is often after the torrent.

The failure of new drugs in the final clinical stage is often because we have chosen the wrong target or other unpredictable reasons. Protein structure prediction does nothing to reduce these two risks, which is why the clinical failure rate of drug development is as high as 85%.

Protein structure prediction is indeed a very thorny problem, but the risks faced in drug research and development are obviously more difficult.

After the Derek Lowe article was published, it also sparked discussion between two groups of readers.

Readers who support him believe that the effects of flexible proteins should indeed be taken into account in the study, because the changes in conformational state need to be understood on a case-by-case basis. Protein-protein and protein-nucleic acid interactions are also important for understanding the system. The structure itself can't solve all the problems, and artificial intelligence still has a long way to go before it can replace the experimental data.

Some readers disagree with Derek Lowe that "good structural prediction will greatly speed up the process of obtaining empirical data sets."

One reader said, "structure-based design will be a limiting factor-in an environment where structures are difficult to obtain." In a world with AlphaFold, this is no longer the case. In addition, you can run AlphaFold again to put a small molecule in and refold the protein around it. Twenty years ago, when I was studying for my doctorate, we used sybyl and autodock to do the same thing-frankly, these software tools were completely rubbish. Traditional drug design is as shaky as a blind person on crutches, and we can now see it through structure-based design. The fact that it (AlphaFold) was not previously an important part of drug design has nothing to do with the discovery of new drugs in the future. "

Some readers believe that structure-based drug design activities greatly help to reduce the failure rate. In the case of lack of experimental structure, AlphaFold combined with other calculation methods such as molecular dynamics simulation is much better than traditional methods.

No matter at home or abroad, scholars have different opinions on AlphaFold, and their views on its impact on drug research and development are also different.

This article by Derek Lowe, which represents the technical experts of mainstream or traditional pharmaceutical companies, is an "instinctive" resistance to new technologies.

This phenomenon is no different from the doctor's complaint about AI when the medical image AI appears. It is essentially a collision and confrontation between two professional backgrounds. However, radiologists are gradually accepting AI to help them find lung nodules.

The answer to this question is also very simple: from what point of view do you evaluate the deep learning technology represented by AlphaFold?

Whose side will you stand on whether AlphaFold can bring about innovative changes in drug research and development?

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.