MaSIF colab

This repo contains a dMaSIF implementation for google colab. For now we only have a dMaSIF site implementation for visualizing protein binding sites. We are working on other implementations, keep an eye on this repo!

Link to the dMaSIF_colab

Available dMaSIF models

Models	Resolution	Dimensions	Layers	Radius	ROC/AUC
dMaSIF site	1.0 A	16	3	12 A	0.87
dMaSIF site	1.0 A	16	3	9 A	0.87
dMaSIF site	0.7 A	16	3	12 A	0.88
dMaSIF site	0.7 A	16	3	9 A	0.88

Models:

dMaSIF site identifies protein–protein interaction sites.
dMaSIF search (COMING SOON!) predicts protein-protein complexes.

Resolution:

The distance between points in the pointcloud in Ångström (lower is higher resolution)

Dimensions:

The number of embedding dimensions

Layers:

Number of convolutional layers

Radius:

Radius to use for the convolution

ROC/AUC:

Performance measure, higher is better.

How to run dMaSIF locally

You can install and run dMaSIF locally throug the original dMaSIF repository.

How to reference this work

Sverrisson, F., Feydy, J., Correia, B. E., & Bronstein, M. M. (2021). Fast end-to-end learning on protein surfaces. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 15272-15281).