ForceSolve

First Release Announcement

We are excited to announce the first release version of ForceSolve, a python code to infer coarse Hamiltonians from forces observed at any fine-grained level of detail.

This software has been under development in the Dr. Beck Lab for the past two years and delivers robust state-of-the-art coarse-graining methods. It has been designed to be one of the easiest (possibly the only), most intuitive, coarse-graining programs currently available for small systems and validation of the force-matching method.

Since new file releases can take a while to be distributed to all the mirrors, you can download the complete project using subversion using the following command:

Feature List

Specification of Arbitrary Molecule Topologies

• Bond specification is done in easily structure-able topology files -- once for each residue type.
• Automatic listing of bond, angle, torsion, and pairwise distance interactions for any given PDB from this topology.
• Automatic assignment of force field atom types using a simple PDB atom:residue name lookup scheme.
• Included script for coarsening of atomic configurations with a site-specification file.

Supported Calculations

• Maximum Likelihood Hamiltonian Inference
• Bayesian Minimum Mean-Squared Error (average) Hamiltonian Estimation
• Coarse-Grained MD using Langevin Dynamics

Molecular Mechanics

• Each forcefield term can be specified by a B-spline of arbitrary range and degree using tabulated coefficients.
• Pre-coded terms exist for pairwise distance, bond, angle, and dihedral interactions.
• Adding novel terms (not in the above list) to the list of calculable interactions requires moderate effort and is well documented.

Included APIs

• Complete implementation of arbitrary polynomial order B-splines and first derivatives.
• Included python library ucgrad handles array and pdb I/O, vector operations such as internal coordinate conversion and generation of rotation matrices, parameter file parsing, and multiple structural superposition.

Unique Properties of our Method

Generated Hamiltonians do not suffer from over-fitting at arbitrarily high number of spline knots, nor do they change with the energy or distance scales of the system under consideration.

Requirements

Runs on any system with Python 2.4 or greater and numpy (numerical python library) installed -- Linux or MS Windows.