Tweets

2023-07-25T01:58:03+00:00 | 🔗

Our XGBoost regressor had a 5-fold cross-validated Pearson’s r of 0.66, just a bit shy of the top model from the JCIM paper (r = 0.77). Bear in mind, our regressor is operating with a limited yet interpretable feature set. #DataAnalysis #AI

2023-07-25T01:58:02+00:00 | 🔗

However, don't use these interpretability plots to directly inform decisions, just like it’s not advisable to directly optimize for PPB. Instead, use them in conjunction with similar plots for other endpoints to understand the overall in vivo efficacy. (https://t.co/sZo6KvuSG2)

2023-07-25T01:58:02+00:00 | 🔗

You can delve deeper by hovering over the graph to see actual counts, then cross-referencing with SHAP values on the x-axis. This shows the predicted additive contribution of each feature to log10 fraction unbound. Remember, +0.3 (log10 scale) indicates a two-fold increase.

2023-07-25T01:58:02+00:00 | 🔗

Our interactive plot display illustrates the relationship between feature, SHAP value, and feature value. We’ve observed trends like that increased NOCount generally raises the model’s predicted % unbound, while reduced rotatable bonds do the same. #DataVisualization #ML

2023-07-25T01:58:01+00:00 | 🔗

We used a XGBoost regressor, trained on Human Plasma Protein binding data (log10 % unbound) from a recent JCIM study. For features we utilized the RDKit’s Lipinski Parameters module to limit to an easily interpretable feature set. (https://t.co/lW2kHx3zgy)

2023-07-25T01:58:01+00:00 | 🔗

Made an interactive, interpretable SHAP visualization on human PPB data presented in Prospective Validation of ML for ADMET, hosted at https://t.co/FhIPQPPyTC (scroll down and hover on plot in desktop, more below) https://t.co/bJjYDgKLoQ

2023-07-25T00:41:35+00:00 | 🔗

RT @sauhaarda: Yesterday, I found out Acrobat lets you copy tables from PDFs. I tried it. It fell short, especially for tables with images…

2023-07-24T19:59:04+00:00 | 🔗

@EricBoittier @andrewwhite01 It won't be AGI, but maybe it could be actually useful :D

2023-07-24T17:00:29+00:00 | 🔗

RT @andyandeggs: This weekend @OlorenAI, we discovered Acrobat has a select table and copy-to-sheets feature. We tried it, but it fell shor…

2023-07-22T22:43:50+00:00 | 🔗

@tjparker @foundersfund @reidhoffman @vkhosla @DavidSacks That list is just PayPal