Two major leaps in AI-powered drug discovery

From AI to Clinical Trials: Fast-tracking Anti-fibrotic Drug Development

The AI-predicted TNIK inhibitor, INS018_055, showcases a rapid transition from AI-based target discovery to clinical validation, offering a new treatment avenue for fibrosis. This study from Insilico Medicine, one of the pioneers in AI for drug discovery, highlights the efficiency of AI in drug discovery, evidenced by the swift 18-month progression from identifying the target to nominating a preclinical candidate, promising a faster route to developing therapies for fibrotic diseases.

For target identification, they used PandaOmics, a cloud-based platform that applies AI and bioinformatics techniques to multimodal omics and biomedical text data for target and biomarker discovery. They even posted a video to show the PandaOmics workflow can identify the TNIK gene as the top targeting candidate in a couple of minutes. (Although we are not sure how long it took them to decide the settings in the workflow and whether the same settings can be directly applied to another disease quickly and successfully.)

For small molecule design, they used Chemistry42, which includes 42 generative models and more than 500 predictive models for scoring which allows researchers to generate small molecules with desired properties from scratch.

For further details, you can access the full article on Nature Biotechnology.

RoseTTAFold All-Atom and RFdiffusionAA: Generalizing Structure Prediction and Design to All Biomolecules

While AlphaFold’s team is still working to expand its capability beyond proteins, RoseTTAFold moved ahead this time. The new RoseTTAFold All-Atom algorithm extends the capabilities of biomolecular modeling by accurately predicting structures of proteins in complexes with small molecules, nucleic acids, and metals. It demonstrates a significant advancement over previous methods, like AlphaFold and RoseTTAFold, by incorporating non-protein molecules and covalent modifications. This breakthrough could significantly impact structural biology, offering insights into complex molecular mechanisms and facilitating innovative drug design.

Note that a large proportion of the training datasets are protein-only data, so it is unclear how accurate and generalizable the model is for other biomolecules and complexes.

For further details, you can access the full article on Science or bioRxiv, and the open-sourced models with weights on GitHub (RoseTTAFold All-Atom and RFdiffusionAA).

If you find the newsletter helpful, please consider:

• 🔊 Sharing the newsletter with other people

• 👍 Upvoting on Product Hunt

• 📧 Sending any feedback, suggestions, and questions by directly replying to this email or writing reviews on Product Hunt

• 🙏 Supporting us with a cup of coffee.

Thanks, and see you next time!