Summary of Improving Generalisability Of 3d Binding Affinity Models in Low Data Regimes, by Julia Buhmann et al.
Improving generalisability of 3D binding affinity models in low data regimes
by Julia Buhmann, Ward Haddadin, Lukáš Pravda, Alan Bilsland, Hagen Triendl
First submitted to arxiv on: 19 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Predicting protein-ligand binding affinity is crucial for computer-aided drug design. Despite advancements, generalizable and performant global binding affinity models remain elusive, especially in low data regimes. The benchmark datasets are not well-suited to assess the generalizability of 3D binding affinity models. Our novel split of the PDBBind dataset minimizes similarity leakage between train and test sets, allowing for a fair comparison between various model architectures. We find that 3D global models outperform protein-specific local models in low data regimes. Moreover, we demonstrate that Graph Neural Networks (GNNs) benefit from three novel contributions: supervised pre-training via quantum mechanical data, unsupervised pre-training via small molecule diffusion, and explicitly modeling hydrogen atoms in the input graph. Our work introduces promising approaches to unlock GNN architectures for binding affinity modelling. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine trying to predict how well a medicine will bind to a protein. This is important because it helps us design new medicines that are effective and safe. So far, we haven’t been able to make good predictions when there’s not much data available. We wanted to see if different types of models would work better or worse in these situations. To test our ideas, we created a special way to split up some existing data so that it’s fair to compare the different models. Our results show that using 3D models can actually be helpful when there’s not much data. We also found that certain techniques, like training the model on quantum mechanics and small molecules, can improve its performance. Overall, our work shows promise for improving our ability to predict how well medicines will bind to proteins. |
Keywords
» Artificial intelligence » Diffusion » Gnn » Supervised » Unsupervised
