Thirty-year drug company veteran Derek Lowe slams AlphaFold: "Pure self-importance" to make drugs based on structural predictions

However, due to the flexibility of the conformation, even the actual data is not fully representative of its accuracy. This is where media reports have exaggerated the impact of the AlphaFold protein structure database on drug development.

In the presence of small molecule ligands, protein structures change and slide, sometimes subtly and sometimes drastically, but AlphaFold cannot yet predict these changes.

It may eventually be possible to find algorithmic solutions to these problems, but so far, there are not enough protein structures that can bind to small molecule ligands. The number we need is very large.

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

Another point, as harsh as it sounds (although it's true): knowledge of protein structures rarely affects the progress of development during drug discovery.

This is because researchers typically run projects based on assays using pure proteins or live cells. The assay data then represent whether the compound meets the investigator's requirements and whether it performs better as new compounds are manufactured.

The structure of the protein may shed light on what compound the researchers make next, but it may not help at all.

Ultimately, it's the real numbers from real biological systems that matter. As drug development projects proceed, these numbers cover pharmacokinetic, metabolic and toxicological assays that can't really be dealt with at the protein structure level.

The rapids are often followed by the final waterfall.

New drugs fail in the final clinical session, often because we have chosen the wrong target or for other unpredictable reasons. Protein structure prediction does nothing to mitigate either risk, which is why the clinical failure rate for drug development is as high as 85%.

Protein structure prediction is indeed a very difficult problem, but the risks faced in drug development are clearly even more difficult.

The publication of this article by Derek Lowe has also sparked discussion between two schools of thought.

Readers who support him argue that the effects of flexible proteins should indeed be taken into account in research, as 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. Structure alone cannot solve all the problems, and AI has a ways to go before it can replace experimental data.

Some readers disagreed with Derek Lowe, saying that "good structure prediction would 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 structure is hard to come by. In a world with AlphaFold, this is no longer the case. Furthermore, it is possible to run AlphaFold again and put a small molecule in and refold the proteins around it.20 Years ago, during my PhD, we used to use sybyl and autodock to do the same thing -- and frankly, these software tools were complete garbage. Traditional drug design faltered like a blind man on crutches; with structure-based design, we can now see. The fact that it (AlphaFold) was not previously an important part of drug design is irrelevant to how new drugs will be discovered in the future."

Some readers have argued that structure-based drug design activities have greatly helped reduce failure rates. Combining AlphaFold in the absence of experimental structure with other computational methods such as molecular dynamics simulations is far better than traditional methods.

AlphaFold has received mixed reviews from scholars both at home and abroad, and opinions on its impact on drug development vary.

This article by Derek Lowe represents the "instinctive" resistance of mainstream or traditional drug company technologists to new technologies.

This phenomenon is no different from the complaints of doctors about medical imaging AI when it appeared, and is essentially a clash between two professional backgrounds. However, radiologists are now accepting AI to help them find lung nodules.

The answer to this question is simple: From what perspective is the deep learning technology represented by AlphaFold valued?

Which side would you take on whether AlphaFold can bring revolutionary changes to the field of drug development?