Structure- and ligand-based multi-target machine learning for virtual screening
Applications are invited for an industrial postdoctoral position under Dr. Valery Polyakov and Dr. Eric Martin at Novartis Institutes for Biomedical Research in Emeryville, California. This interdisciplinary postdoctoral opportunity in Computational Sciences will work at the interface of structure-based drug design and machine learning.
Inexpensive computational virtual screens by docking have not had the accuracy needed to replace expensive and time-consuming experimental high-throughput screens. QSAR can be accurate for compounds similar to the training set, but not for interesting novel chemical matter.
Our lab is using multi-target machine learning to overcome these limitations.
AutoShim creates accurate, target customized scoring functions by adjusting the weights of pharmacophore shims in a protein binding site to reproduce several hundred training IC50s.1 Kinase Ensemble Surrogate AutoShim pre-docks the screening collection into an ensemble of 8 diverse representative kinases, which can then be “shimmed” to predict the activities of the entire collection on all other kinase with training data, without requiring further docking or even protein structures.2 Profile-QSAR is a 2D ligand substructure-based method that expands the domain of applicability of empirical models to the entire company archive by using predicted activity from thousands of conventional single-assay 2D QSAR models as compound descriptors for modeling new assays.3,4
This project will expand on these methodologies as well as developing new approaches.
The qualified candidate will be a highly motivated, creative and independent computational chemist with expertise in modern machine learning methods including multi-target deep neural networks.
The candidate should be a competent programmer familiar both with procedural languages, ideally python, as well as machine learning packages including scikit learn or R and tensorflow. Familiarity with docking will be important, and 3D QSAR will be a plus.
The position will start January 2018. Please apply through http://postdoc.nibr.com/eric-martin.html
1. Martin, E. J.; Sullivan, D. C., AutoShim: empirically corrected scoring functions for quantitative docking with a crystal structure and IC50 training data. J. Chem. Inf. Model. 2008, 48, 861-872.
2. Martin, E. J.; Sullivan, D. C., Surrogate AutoShim: Predocking into a Universal Ensemble Kinase Receptor for Three Dimensional Activity Prediction, Very Quickly, without a Crystal Structure. J. Chem. Inf. Model. 2008, 48, 873-881.
3. Martin, E.; Mukherjee, P.; Sullivan, D.; Jansen, J., Profile-QSAR: a novel meta-QSAR method that combines activities across the kinase family to accurately predict affinity, selectivity, and cellular activity. J Chem Inf Model 2011, 51, 1942-56.
4. Martin E.J.; Polyakov, V.R.; Tian, L.; Perez, R.C., Profile-QSAR 2.0: Kinase Virtual Screening Accuracy Comparable to Four-Concentration IC50s for Realistically Novel Compounds. J Chem Inf Model 2017, DOI: 10.1021/acs.jcim.7b00166.