International Journal of Science, Engineering and TechnologyAuthors: Kale Anjali Rajesh, Kale Shanur Dnyaneshwar, Kambale Swapnil Sanjay, Kamble Aryan Ravindra, Kamble Riddhi Shrikrishna, Kamble Yash Ajay, Karande Priya Abaji, Katkar Pratiksha Dnyandeo
Abstract: The rapid advancement of artificial intelligence (AI) and deep learning has transformed molecular property prediction in areas such as drug discovery, material science, computational chemistry, and biotechnology. Accurate prediction of molecular properties reduces the dependence on expensive and time-consuming laboratory experiments while accelerating scientific innovation. The reviewed work, MolBench: A Benchmark of AI Models for Molecular Property Prediction, introduces a benchmark framework designed to evaluate AI models across multiple molecular prediction tasks using multidimensional metrics and standardized datasets. This review paper provides a detailed and original analysis of the MolBench framework, including its objectives, datasets, methodologies, evaluation metrics, experimental findings, strengths, limitations, and future implications. The paper critically examines the benchmark’s contribution to molecular machine learning by comparing traditional machine learning methods, graph neural networks, and self-supervised pre-trained models. Furthermore, it discusses the significance of benchmark-driven research in ensuring fairness, reproducibility, and scientific progress in AI-based molecular property prediction.
DOI: http://doi.org/10.5281/zenodo.20547029
