Minor Release (v0.1.1)

This is a minor release of gym-lattice, an HP 2D-Lattice environment with a Gym-like API for the protein folding problem. This release focuses on fixing the compute_free_energy() function and improving the documentation in the README.

Fixing the compute_free_energy() function

As it turns out, the previous implementation of the free energy computation includes the adjacent H-H pairs even if they are neighbors in the sequence. The standard computation should not include that. This release fixes the problem and updated the expected values during testing.

Better documentation in the README.

Instead of manually building assets in a graphical software, all images are now created using LaTeX. We use the standalone package to generate PNG files from our TeX files. This gives a cleaner version of our images, and hopefully a better representation of our problem space.

Major Release (v0.1.0)

Gym-lattice is an HP 2D Lattice Environment with a Gym-like API for the protein folding problem.

This is a Python library that formulates Lau and Dill's (1989) hydrophobic-polar two-dimensional lattice model as a reinforcement learning problem. It follows OpenAI Gym's API, easing integration for reinforcement learning solutions.

Features

• OpenAI Integration: uses Gym's API to ease compatibility to reinforcement learning solutions.
• Lattice 2D Environment: implements Dill and Lau's two-dimensional lattice as an RL problem.
• Command-line rendering environment: the method render() draws the chain on the command-line.

Additionally, there is an option to set the penalty parameters for training the agent, this includes:

• Collision penalty (collision_penalty): accounts for the time whenever the agent decides to assign a molecule to an already-occupied space; and
• Trap penalty (trap_penalty): induces heavy deductions whenever the agent traps itself and cannot accomplish the task anymore.

Tests

• Error-handling: all public-facing methods now use Exceptions instead of asserts when catching errors.
• Testing with pytest and tox: unit-testing is being done with pytest and tox.

Initial Release (v0.0.1)

This is a pre-release for the gym-lattice environment that provides an easy API for protein folding in a reinforcement learning setting. This is currently a proof-of-concept so please expect possible changes in the internal API (except for the basic gym methods).