Coalesix

🦕 Before Insilico Medicine, Recursion, Atomwise and many others, CoalesiX, launched in 2006, used another form of AI, Evolutionary Computation (EC), to turbocharge the ability of medicinal chemists to explore a space of small molecules. EC was combined with active learning, whereby the medicinal chemist provided feedback to the algorithm by rating, modifying molecules generated by EC and fine tuning the fitness function (or loss function in ML/AI parlance). The algorithm could access computational models from vendors such as Schrödinger to evaluate the loss function. The value was in pairing a generative algorithm and computational SAR models with a human expert, applying the experience of the chemist where it matters. Our first pilot was so successful that the then-head of CompChem at Eli Lilly and Company took his retirement early and joined the newly formed company.

☠️ CoalesiX failed. Not because it was not a great tool, not because it did not produce spectacular results. It failed because selling software to big pharma, or any biotech for that matter, is really painful, with a multi-year sales cycle. When we realized that, we tried to pivot to using our technology to develop our own drug candidates because we thought we had a real competitive advantage. No one really bit, it was too early and the concept that machine learning could provide a significant advantage was alien to investors. It sucks to be early.

As for Schrödinger, founded in 1990, it took them 2 years to get their first software sale. 34 years after its founding, the company had about $160m in software/saas revenue (75% of its total evenue) in 2023. That tells you how bad that business is. No wonder then, that they decided a few years back to also develop their own assets, or partner, including their unique and fascinating deal with Nimbus Therapeutics (with a $147.3 Million distribution from the sale of Nimbus' TYK2 inhibitor to Takeda).

Schrödinger's market cap reflects the toughness of the software side: $1.6bn, which, although not terrible, is a fraction of Recursion's or other "AI-first" public biotech companies. Insilico Medicine, led by the fearless and never-sleeping Alex Zhavoronkov, is one of the most promising companies with an end-to-end stack, fingers crossed for phase 3 success, which would lead to a multi-billion valuation.

💡 What is interesting to me, 15 years after folding the company, is that the backbone of the approach -hashtag#genAI + active learning, is still not only very much relevant but central to any hashtag#AI4DD approach. I am hoping to jump back into that pond with a clear clinical goal, modern tools and enlightened investors 😎