Summary of Transfer Learning For Molecular Property Predictions From Small Data Sets, by Thorren Kirschbaum and Annika Bande
Transfer Learning for Molecular Property Predictions from Small Data Sets
by Thorren Kirschbaum, Annika Bande
First submitted to arxiv on: 20 Apr 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Chemical Physics (physics.chem-ph)
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 Machine learning has become a valuable tool in chemistry for predicting molecular properties, such as in high-throughput screening applications. However, many studies rely on small datasets, making it difficult to implement powerful deep learning architectures like message passing neural networks (MPNNs). This study benchmarks common machine learning models on two small datasets and finds that MPNNs, specifically PaiNN, achieve the best results when combined with SOAP molecular descriptors and gradient boosting regression trees. To further improve predictive capabilities, a transfer learning strategy is presented, which uses large pre-training datasets to fine-tune models on smaller target datasets. This study also explores the impact of pre-training data set size on final training results. |
Low | GrooveSquid.com (original content) | Low Difficulty Summary Machine learning is helping chemists predict important properties of molecules without expensive experiments or calculations. The problem is that many studies use small amounts of data, making it hard to use powerful AI techniques like deep learning. A new study looks at how well different AI models work on two small datasets and finds that a special type of deep learning called message passing neural networks (MPNNs) does best when combined with other methods. To make these models even better, the researchers developed a way to use larger amounts of data to “train” the models before using them for specific predictions. This study also shows how using more or less pre-training data can affect the final results. |
Keywords
» Artificial intelligence » Boosting » Deep learning » Machine learning » Regression » Transfer learning