Patrick Walters

951 posts

Patrick Walters banner
Patrick Walters

Patrick Walters

@wpwalters

cheminformatics, machine learning, drug discovery, opinions

Boston เข้าร่วม Haziran 2009
375 กำลังติดตาม6.3K ผู้ติดตาม
Patrick Walters
Patrick Walters@wpwalters·
@andrewwhite01 Nice, you should ask kosmos to submit to the OpenADMET/ExpansionRx blind challenge.
English
0
0
4
342
Andrew White 🐦‍⬛
Andrew White 🐦‍⬛@andrewwhite01·
Then it found a lightgbm model was better. Rather than throw out the GNN, it used the embeddings from it as a new feature in the lightgbm model and it found it improved fit. 2/2
Andrew White 🐦‍⬛ tweet mediaAndrew White 🐦‍⬛ tweet media
English
6
0
34
3.3K
Andrew White 🐦‍⬛
Andrew White 🐦‍⬛@andrewwhite01·
I asked kosmos to train a model on ADMET data. It built a multitask graph neural network, trained it, noticed it had slow convergence, inspected the gradient norms of the task heads, and then transformed the data to Z-scores to fix it. Cool to watch it work 1/2
Andrew White 🐦‍⬛ tweet mediaAndrew White 🐦‍⬛ tweet media
English
9
26
219
26.8K
Daniel Dale
Daniel Dale@ddale8·
Here's a fact check of 18 of the false claims President Trump made in his lie-filled interview on "60 Minutes" last night: cnn.com/2025/11/03/pol…
English
227
2.5K
6.3K
414.7K
David Mobley
David Mobley@davidlmobley·
Are cofolding models moving beyond memorization? My research group's journal club is discussing a preprint on it this week, and it's worth a look. biorxiv.org/content/10.110…
English
1
0
13
892
Patrick Walters
Patrick Walters@wpwalters·
@Azhar__Shaikh Given my chemistry background, I have my own biases, but I believe that domain knowledge is essential for anyone involved in ML for drug discovery.
English
0
0
0
60
Azhar Shaikh
Azhar Shaikh @Azhar__Shaikh·
@wpwalters As someone with ML/DL background is it easier to get into drug discovery(learn bio stuff on the go) or is having advanced bioinformatics/biomedical degrees a hard requirement
English
1
0
10
108
Patrick Walters
Patrick Walters@wpwalters·
Excellent new paper (with code) by my former colleagues Steven Kearnes and Patrick Riley describing a procedure for associating confidence levels with regression model predictions in drug discovery. pubs.acs.org/doi/10.1021/ac…
English
0
9
63
5.1K
Patrick Walters
Patrick Walters@wpwalters·
I'm not happy with the way this all went down. I have tremendous respect for Gabriele and the rest of the DiffDock authors. Their work has broken new ground and helped advance machine learning in drug discovery. If I had to do it over again, I'd do things differently.
English
3
2
25
6.1K
Patrick Walters
Patrick Walters@wpwalters·
Many of you have seen Gabriele Corso's response to our recent preprint. If you haven't, it's here. lnkd.in/eNvvsRPx
English
2
4
31
11.8K
Patrick Walters
Patrick Walters@wpwalters·
@DdelAlamo @prof_ajay_jain Time splits with the pdb are a profoundly bad idea. Groups often deposit a novel structure and then deposit the structures of analogs a couple of years later. Efforts like Plinder are addressing this.
English
1
4
18
1.7K
Diego del Alamo
Diego del Alamo@DdelAlamo·
@wpwalters @prof_ajay_jain This is some pretty strong language in the conclusion - are the broader problems outlined here unique to DiffDock? The temporal splits used here are common in other fields, while previous-generation software gets misused all the time in scientific benchmarks
English
2
0
8
2K
Patrick Walters
Patrick Walters@wpwalters·
I'm thrilled to announce a new preprint describing collaborative work with @prof_ajay_jain and Ann Cleves Jain, "Deep-Learning Based Docking Methods: Fair Comparisons to Conventional Docking Workflows". arxiv.org/abs/2412.02889
Patrick Walters tweet media
English
5
25
106
23.8K