Semi-supervised learning boosts QSAR models with selective unlabeled data

Researchers have explored the effectiveness of semi-supervised learning (SSL) for Quantitative Structure-Activity Relationship (QSAR) modeling, particularly when using large chemical repositories like ChEMBL. Their study, involving 10 QSAR datasets and over 400,000 unlabeled molecules, evaluated three SSL algorithms and four data selection strategies. The findings indicate that unlabeled data can indeed enhance QSAR models, with certain configurations outperforming supervised baselines. Notably, selecting a small number of chemically diverse molecules proved more beneficial than simply increasing the volume of unlabeled data, suggesting that the quality and composition of the unlabeled set are critical for successful SSL in this domain. AI

IMPACT This research suggests that careful selection of unlabeled data can significantly improve the performance of QSAR models, potentially reducing the need for extensive and costly experimental labeling in drug discovery and chemical research.

RANK_REASON Academic paper detailing a new methodology for machine learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• ChEMBL
• Co-training
• furthest-neighbor sampling
• label propagation algorithm
• Mixed-Sampling Algorithm Combining Cluster Boundary Movement and Adaptive Synthesis
• nearest-neighbor sampling
• QSAR
• random sampling
• semi-supervised learning

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →