Learning to Make Chemical Predictions: the Interplay of Feature\n Representation, Data, and Machine Learning Algorithms

2020/02/28 by Mojtaba Haghighatlari, Haghighatlari, Mojtaba, Jie Li +9
Computer Science · Biochemistry, Genetics and Molecular Biology · Chemistry · #Computational Drug Discovery Methods #Metabolomics and Mass Spectrometry Studies #Various Chemistry Research Topics

paper · pdf · doi:10.48550/arxiv.2003.00157

Abstract

Recently supervised machine learning has been ascending in providing new\npredictive approaches for chemical, biological and materials sciences\napplications. In this Perspective we focus on the interplay of machine learning\nalgorithm with the chemically motivated descriptors and the size and type of\ndata sets needed for molecular property prediction. Using Nuclear Magnetic\nResonance chemical shift prediction as an example, we demonstrate that success\nis predicated on the choice of feature extracted or real-space representations\nof chemical structures, whether the molecular property data is abundant and/or\nexperimentally or computationally derived, and how these together will\ninfluence the correct choice of popular machine learning algorithms drawn from\ndeep learning, random forests, or kernel methods.\n

Citations

• The perceptron: A probabilistic model for information storage and organization in the brain.
• Backpropagation Applied to Handwritten Zip Code Recognition
• SchNet – A deep learning architecture for molecules and materials

Related

• IMPRESSION -- Prediction of NMR Parameters for 3-dimensional chemical\n structures using Machine Learning with near quantum chemical accuracy
• Machine Learning, Quantum Mechanics, and Chemical Compound Space
• Advancements in Molecular Property Prediction: A Survey of Single and Multimodal Approaches
• Combining Machine Learning and Computational Chemistry for Predictive Insights Into Chemical Systems
• Highly Accurate Prediction of NMR Chemical Shifts from Low-Level Quantum Mechanics Calculations Using Machine Learning