Note: I am in San Francisco right now and, in an extraordinary coincidence, I stumbled across two of the people whose work I mention in this article! Very grateful to John Bradshaw for chatting about reaction prediction and Gina El Nesr for chatting about molecular simulation.
A second note: while here in SF, I will be co-hosting an event on Friday, Jan 16th, from 6-9pm, w/ Tamarind Bio! It will be at Southern Pacific Brewing, here is the link to the invite. You should come by!
Introduction
There are two memories that I have to imagine are particularly heartwarming for any parent. One, seeing their child for the first time, and two, gleefully showing photographs of that child to an older version of that child, shouting, look how small you used to be! So small! Do you know how hard I worked to take care of you? You were so difficult! But it’s okay, because you were so, so tiny.
I will do something similar to this today. This blog has been operating for the exceptionally long period of 1.7~ years, which means I finally have blog posts that I wrote back in 2024 to resurface, dust off, and proudly present back to you, giving you an update on how things have shifted in the 1~ years since they were written.
I will do this for three articles, back when my cover images were stranger:
It’s fun looking back at these three in particular, because they all feel intellectually significant. All of them were, essentially, predictions of where the future in a specific subfield of bio-ML may go. The first was the first time I’d ever seriously engaged in the small-molecule design space, the second for the molecular dynamics space, and the third for what are durable startup plays. Each one required multiple conversations with multiple people, many of whom I’d talked to the first time ever, and some I continue talking to today. Nostalgic!
But why do this at all? It would be easy to write confidently about the future and then quietly memory-hole the predictions when they don’t pan out, which, to be clear, there’s nothing wrong with and I likely will do many times. This is a blog, nobody cares that much. Still, it is worth doing this purely because it forces me to wrap my head around what has changed since I last covered something, not merely everything that is new and exciting. This is a little boring, but it does feel an important muscle to flex for the same reason that it is important to do your A-B-C’s every few months; just making sure you’re still capable of accomplishing the fundamentals.
As for format: for each article, I’ll briefly recap the original thesis, look at what has actually happened since, and render some kind of verdict as to what went right/wrong. I’ll also attach a tl;dr at the top of each section.
Generative ML in chemistry is bottlenecked by synthesis
... continue reading