Summary of Pretraining Graph Transformers with Atom-in-a-molecule Quantum Properties For Improved Admet Modeling, by Alessio Fallani et al.
Pretraining Graph Transformers with Atom-in-a-Molecule Quantum Properties for Improved ADMET Modeling
by Alessio Fallani, Ramil Nugmanov, Jose Arjona-Medina, Jörg Kurt Wegner, Alexandre Tkatchenko, Kostiantyn Chernichenko
First submitted to arxiv on: 10 Oct 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 This paper evaluates the impact of pretraining Graph Transformer architectures on atom-level quantum-mechanical features for modeling ADMET properties of drug-like compounds. Three pretraining strategies are compared: one based on molecular quantum properties, another using self-supervised atom masking, and a third utilizing atomic quantum mechanical properties. After fine-tuning on Therapeutic Data Commons ADMET datasets, the models pretrained with atomic quantum mechanical properties generally produce better results. The study analyzes latent representations, finding that supervised strategies preserve pretraining information after fine-tuning and exhibit different trends in latent expressivity across layers. Additionally, the paper demonstrates that models pretrained on atomic quantum mechanical properties capture low-frequency laplacian eigenmodes of input graphs via attention weights, producing better representations of atomic environments within molecules. The analysis is applied to a large non-public dataset for microsomal clearance, illustrating generalizability and highlighting performance differences between model types. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper explores how to improve computer models that predict how well drugs work in the body. It compares three ways of training these models before they’re used for predictions. The researchers found that one method, using information about atoms in molecules, works best. They also looked at what the models are learning and found that this method produces more useful information. The study shows that different methods can produce similar results on small datasets but have different strengths when applied to larger datasets. |
Keywords
» Artificial intelligence » Attention » Fine tuning » Pretraining » Self supervised » Supervised » Transformer
