cg-flatland 0.0.1

Easier development of polymer structure prediction tools.

Flatland

Easier development of polymer structure prediction tools

The development and debugging of modern neural network approaches for protein structure prediction has the potential to be a somewhat complicated task. Here we provide a framework, Flatland, for simulating training datasets of the form used by current state-of-the-art systems. The central task performed is to evolve a population of polymer sequences, predict their folded structure, produce distance matrix and bond angle tensors, and calculate spectra of per-polymer compound or other polymer interaction likelihoods (emulating protein-protein and compound-protein interaction datasets). This framework intentionally performs such simulations in a highly simplified manner that does not accurately model the real biological processes. Relatedly, it provides the means to simplify the problem to an arbitrary extent - including permitting the reduction from three to two dimensions, reducing the amino acid alphabet from twenty-one to three, and simplifying characteristics of the evolutionary optimization. We provide the framework together with a collection of starting-point notebooks for training neural network models to learn from such data by way of an open source repository to kick-start the development efforts of those interested in contributing to solving this vitally important problem.

Installation

You can install the last released version of Flatland via pip

pip install cg-flatland

Or install the most recent development version of Flatland using git

git clone https://github.com/cayley-group/flatland
pip install -e flatland

Demonstrations

We're putting together various demonstration notebooks which you'll be able to access via the following:

• Data Generation
• Structure Solver Example