How is Artificial Intelligence impacting the design and discovery of new compounds and medications?

Given the rapid advancements in artificial intelligence and machine learning methodologies over the past decade, how are these computational tools being integrated into the realm of chemistry, specifically in the design and discovery of novel compounds and medications?

Moreover, how do these AI-driven techniques compare in terms of efficiency, accuracy, and innovation to traditional human-driven research methods? And, what potential challenges or ethical considerations arise when relying heavily on AI for such pivotal and impactful scientific endeavors?

  • 2
    $\begingroup$ In context of pre-synthesis computational screening of potential drug candidates, AI starts to compete with quantum chemistry computations. $\endgroup$
    – Poutnik
    Sep 1, 2023 at 9:27
  • 5
    $\begingroup$ Seems a bit broad in scope. You may want to find a more specific question that will not require an essay to answer. Questions regarding ethics are not generally for this site but you may look for a more specific question. Potential challenges sounds a bit broad and fuzzy, do you have any particular examples in mind? $\endgroup$
    – Buck Thorn
    Sep 1, 2023 at 10:46
  • 1
    $\begingroup$ @Poutnik I'm curious, do you have an example or news article for that? I know that ML approaches have been used and developed for pre-screening purposes in the last millennia. I wonder where the AI came in. As far as I know, this has never extended (practically) to the realm of QC computations as it is too costly. Many semi-empiric and dft-substituting approaches have been developed to tackle the inefficiencies. So I'm really curious where this is heading. Given the verbosity though, this is better discussed in Chemistry Chat. $\endgroup$ Sep 2, 2023 at 11:39
  • $\begingroup$ @Martin Hi Martin, I posted the follow up post in Periodic table chat But no special new info about AI in O. chemistry. $\endgroup$
    – Poutnik
    Sep 25, 2023 at 9:10

1 Answer 1


Can you narrow the question? Perhaps cheminformatics applying e.g. PCA for decades to recognize the most important contributions to a function (see the primer by Sidou and Borges, for example) is less visible to you than protein folding (keyword alphafold) which equally appeared in the press not specialized on chemistry. There is at least one journal dedicated to the field, Journal of Cheminformatics. With Craig Plot by Ertl, for example, data sets can be analyzed for common substituents. Perhaps your company wants to commercialize a product in the same niche as your competitor, but to circumnavigate their patent, the structure of your active molecule has to be sufficiently structurally different on paper to merit a separate IP protection while the ADMET properties are still similar enough (or better) to justify the investment.

AI can help in the identification of plausible routes of synthesis. Formalized retrosynthesis e.g. by Corey has been a field of application for long; digitized databases of known reactions and publications crawled by OCR ease to build/improve tools like the open source AiZynthFinder (Reymond group, University of Berne), ASKCOS at MIT, or closed source IBM RXN for Chemistry. (Too many to name them here, Wikipedia compiled a list.) It equally extends to virtual catalyst screening, too (e.g., Yang et al.). Similar to AI in ChatGPT, it can become a very helpful guide / source of inspiration to follow a route you are less familiar with, because your working group tackles a synthesis of motif A preferentially / always by a special method developed locally. The potential danger to students (and users/workflows in general) is relying too much on these tools. For one, what happens if you change your current group for one which does not have a subscription to this infrastructure [anymore]? What are quality and coverage of raw data the AI accessed, and their digestion / model building (else AI is perhaps not about artificial intelligence, but an artificial idiot)?

Note, Artificial Intelligence Applications in Chemistry is one recognized keyword on the landing page for journals by The American Chemical Society.* As by today, there are 7741 publications by this publisher alone. The search by this link can be narrowed further e.g. by the type of publication (article, review, perspective), or/and topic (based on your question, biological chemistry, medicinal chemistry, organic chemistry possibly the more interesting sections for you). With artificial intelligence alone (link), there are 10066 (and counting). And ACS is only one (of the larger) publishers relevant to chemistry.

* one of the analogues about chemometrics.

(1) Sidou, L. F.; Borges, E. M. Teaching Principal Component Analysis Using a Free and Open Source Software Program and Exercises Applying PCA to Real-World Examples. J. Chem. Educ. 2020, 97, 1666–1676. doi 10.1021/acs.jchemed.9b00924.

(2) Ertl, P. Craig Plot 2.0: An Interactive Navigation in the Substituent Bioisosteric Space. J. Cheminform 2020, 12, 8. doi 10.1186/s13321-020-0412-1.

(3) Yang, W.; Fidelis, T. T.; Sun, W.-H. Machine Learning in Catalysis, From Proposal to Practicing. ACS Omega 2020, 5, 83–88. doi 10.1021/acsomega.9b03673 / entry on pubchem.