New preprint on pre-trained models for Bayesian optimization (BO) of sequences! We show LLMs trained on protein seqs can replace Gaussian processes in BO. Examples: BO of peptide inhibitors with AlphaFold and iterative design of proteins. 1/6 biorxiv.org/content/10.110…

Use ensembles of pretrained protein LMs + reparametrization trick for Bayesian optimization of protein sequences. Ziyue Yang 🐣 kat !! Andrew White 🐦‍⬛ biorxiv.org/content/10.110…

How to avoid burnout. Get 7 types of rest:

I am happy to share lecture notes, videos, and jupyter notebooks I developed for my #Compchem class at Dartmouth Chemistry: "CHEM101.6: Computational Methods In Chemistry & Biophysics" We also use these resources as tutorials for new students in our group. github.com/paulrobustelli…

I have had such great colleagues and the coolest boss at this lab! They are such a supportive and empowering bunch.. going to miss all of you too ♥️ Geemi Wellawatte Andrew White 🐦‍⬛ TheRealAmoo 🤖🧪 Ziyue Yang 🐣 Sam Cox Shane Michtavy @Kyam888

This is what I worked hard for... Andrew White 🐦‍⬛ saying this means so much 🥺 Thank you for all the guidance, encouragement, mentorship and patience through the last few years!! You are the BEST boss everrrr !!!

Solved a problem that I've been thinking about since John Pinkus and me did a homework in 2008 - automatically naming fragments of molecules. Will be in next version of exmol!

Congratulations to our very own Dr. Gandhi! Heta Gandhi Hopefully I'll be like you when I grow up 💕

Grad school journey ends! Phinally Done ✅ Thanks Andrew White 🐦‍⬛ for all the wonderful projects and research opportunities and being an amazing PhD mentor! Going to miss the lab very much!

Why do molecules smell? We create explanations for why molecules have one of 112 scents. The explanations are distilled into natural language. We're excited about data -> model -> explanation of mols. Work by Aditi (looking for PhD program!), Heta Gandhi, Geemi Wellawatte 1/3

New paper giving our perspective for how to explain deep learning models in chemistry. We hope the simplicity and utility is clear, and that more people will start providing explanations for predictions! By Geemi Wellawatte, Heta Gandhi, Aditi chemrxiv.org/engage/chemrxi…

My research hit the 100 citations mark! 🥳 Academic Chatter™

So proud of this young scientist! We’ve worked together closely and Geemi Wellawatte has been an excellent collaborator!

[JCTC] [ASAP] A Perspective on Explanations of Molecular Prediction Models dlvr.it/SlZgtt

A great beginning to the week. Latest publication is now out! Heta Gandhi Andrew White 🐦‍⬛ A Perspective on Explanations of Molecular Prediction Models pubs.acs.org/doi/10.1021/ac…

Three years ago, I started working on an easy-to-use tool for interpretable machine learning in science. I wanted it to do for symbolic regression what Theano did for deep learning. Today, I am beyond excited to share with you the paper describing it! arxiv.org/abs/2305.01582 1.

Symbolic regression (SR) finds analytic equations from data, like Newton's law of gravity. We show how to use SR to design experiments: which experiment falsifies as many hypothesized equations at once. Work by Jorge Medina 1/4 arxiv.org/abs/2305.10379

Not able to produce “favorable results” in the lab? You are a “loser”. Stay in the shadows of those who can bring cool data to the Big Prof. Sounds familiar? It's not only about this case at Stanford. It's about academic culture in general: - Keep pushing for metrics.

Checkout our latest in-browser ML website for predicting molecular properties! @Kyam888 and I implemented deep ensembled neural networks to accurately predict solubility w/error bars as you type - mol.dev Paper: arxiv.org/abs/2307.05318