Summary of Explainable Molecular Property Prediction: Aligning Chemical Concepts with Predictions Via Language Models, by Zhenzhong Wang et al.
Explainable Molecular Property Prediction: Aligning Chemical Concepts with Predictions via Language Models
by Zhenzhong Wang, Zehui Lin, Wanyu Lin, Ming Yang, Minggang Zeng, Kay Chen Tan
First submitted to arxiv on: 25 May 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 The paper presents a novel framework, called Lamole, for providing explainable molecular property predictions. Lamole is based on transformer-based language models, but it also offers chemically meaningful explanations of its predictions. The approach uses Group SELFIES, a string-based representation of molecules, as input tokens to pretrain and fine-tune the model. To quantify the impact of each substructure on the output, the paper proposes combining self-attention weights and gradients. Additionally, the authors develop a marginal loss function to optimize explanations that align with chemists’ annotations. The approach is tested on six mutagenicity datasets and one hepatotoxicity dataset, achieving comparable classification accuracy and improving explanation accuracy by up to 14.3%. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper creates a new way to predict properties of molecules, like whether they are safe for use in medicine or not. This prediction needs to be explained in a way that makes sense to chemists who understand the structure of molecules. The approach uses special strings called Group SELFIES that represent the molecule’s structure. It also combines two types of information, attention and gradients, to figure out how different parts of the molecule affect the predicted property. To make sure the explanations are accurate and helpful, the authors developed a way to adjust the model so its predictions align with what chemists would expect. |
Keywords
» Artificial intelligence » Attention » Classification » Loss function » Self attention » Transformer
