SFcalculator

This is the code repo related to short paper Towards automated crystallographic structure refinement with a differentiable pipeline on Machine Learning in Structural Biology Workshop at NeurIPS 2022.

Structure Factor Calculator implemented in tensorflow2, pytorch and jax.

A differentiable pipeline connecting the protein atomic models and experimental structure factors, featuring a differentiable bulk solvent correction.

The symmetry-related nitty-gritty in both real space and reciprocal space are included.

A more detailed manuscript is in preparation.

Source codes

Source codes in three popular deep learning frameworks are provided in the following submodule repositories:

1. SFcalculator_torch, pytorch implementation.

2. SFcalculator_jax, jax implementation.

3. SFcalculator_tf, tensorflow2 implementation.

The pytorch version is currently in active development, making it several versions ahead and the preferred choice.

Authors

Minhuan Li, [email protected]

Doeke R. Hekstra, [email protected]

Installation

Pytorch verision

1. Create a python environment with package manager you like (mambaforge recommended).

2. Install Pytorch

3. Install SFcalculator-torch

   pip install SFcalculator-torch

Jax verision

1. Create a python environment with package manager you like (mambaforge recommended).

2. Install Jax

3. Install SFcalculator-jax

   pip install SFcalculator-jax

Tensorflow 2 verision

1. Create a python environment with package manager you like (mambaforge recommended).

2. Install Tensorflow2

3. Install SFcalculator-tf:

   pip install SFcalculator-tf