Tools
There is an overwhelming variety of tools that have been used to predict the binding affinity between a small molecule and a protein. We have gone through the literature and open source code repositories to provide a selection of tools that are ready for you to run them on PLEx.
Note: All models we provide are research-grade software and are provided "as-is". No model for this task has yet been demonstrated to generalise well enough to be an alternative to laboratory experiments. When providing a tool we make use of existing, often academic, contributions. Please give credit to the creators of open-source work. We are standing on the shoulder of giants.
At this point in time we are focused on pose prediction. Stay tuned for integrated scoring functions.
Small Molecule Binding Affinity Prediction
In the chart below we give you an overview of the models available. We would generally recommend to optimise for a high model accuracy when predicting small molecule binding affinity.
Mini: Equibind
Equibind is a very fast machine learning model that can approximate the docking pose of a small molecule and a protein. The model is less accurate than baseline methods, such as Gnina (below), but orders of magnitute faster.
Please cite
Stärk, H., Ganea, O.-E., Pattanaik, L., Barzilay, R., & Jaakkola, T. (2022). EquiBind: Geometric Deep Learning for Drug Binding Structure Prediction. http://arxiv.org/abs/2202.05146
Base: Gnina
Gnina is a physics-inspired docking and scoring tool that uses a convolutional neural network to score poses. Gnina is an implementation of Smina, which itself is a fork of Vina.
Please cite
A McNutt, P Francoeur, R Aggarwal, T Masuda, R Meli, M Ragoza, J Sunseri, DR Koes. J. (2021). GNINA 1.0: Molecular docking with deep learning https://chemrxiv.org/engage/chemrxiv/article-details/60c753ebbb8c1a1a9d3dc142
Standard: Diffdock
Diffdock is a machine learning model that predicts the docking pose of a small molecule and a protein. It frames molecular docking as a generative problem. When it comes to predicting the 3D configuration of a small molecule and protein within a standard margin of error (2µm), Diffdock has a 38% top-1 success rate and is faster and more accurate than existing publicly available tools.
Please cite
Corso, G., Stärk, H., Jing, B., Barzilay, R., & Jaakkola, T. (2022). DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking. http://arxiv.org/abs/2210.01776