DI-bioseq

Introduction

DI-bioseq is an open-source Decision Intelligence platform for biological sequence prediction and searching. DI-bioseq provides a Reinforcement Learning pipeline for biological sequence searching, including DNA, RNA, and amino acid sequences. DI-bioseq uses DI-engine, an RL platform to build searching methods. DI-bioseq is an application platform under OpenDILab. DI-bioseq was developed referencing FLEXS on biological environments and modules.

Outline

• Introduction
• Installation
• Features
• Quick Start
• Contributing
• License
• Citation

Introduction

Biological sequence searching problems aims to find a sequence with a fixed length of these elements that has the best characteristics that humans desired, represented by a score. The search usually starts from a series of known sequences whose scores are measured via biological experiments.

Considering the high cost and time consumption of real biological experiments required for verifying the results, the searching procedure is expected to be high-efficient, with fewer scores referred to from biological experiments, rather than proposing lots of candidates and iteration many rounds to find the best one.

Biological sequence searching pipelines often consists of the following modules:

• A biological ground-truth data/simulator that can get the fitness score of a proposed sequence, always called Landscape
• A prediction Model that attempts to fit the ground-truth landscape from known score
• An Encoder to convert a sequence into numbers
• An Engine which is the core method to find new sequence proposals from known ones and scores

Installation

DI-bioseq together with DI-engine can be easily installed from source via pip.

git clone https://github.com/opendilab/DI-bioseq.git
cd di-bioseq
pip install -e . --user

Features

DI-bioseq currently supports the following searching modules.

Landscape	Engine	Model	Encoder
GB1	Random	Linear Regression	Onehot
TF-Binding	Off-policy PPO(onehot only)	Random Forest Regression	Georgiev(protein only)
	AdaLead

Quick Start

DI-bioseq provides a standard searching pipeline and entry that can be configured by arguments.

usage: main.py [-h] [--landscape LANDSCAPE] [--engine ENGINE] [--model MODEL]
               [--encoder ENCODER] [--score SCORE] [--predict PREDICT]
               [--round ROUND] [--logdir LOGDIR]

optional arguments:
  -h, --help            show this help message and exit
  --landscape LANDSCAPE
                        landscape type
  --engine ENGINE       searching engine type
  --model MODEL         prediction model type
  --encoder ENCODER     encoder type
  --score SCORE         max score in landscape each round
  --predict PREDICT     max prediction in model each round
  --round ROUND         searching rounds
  --logdir LOGDIR       result logging folder path

Join and Contribute

We appreciate all contributions to improve DI-bioseq, both algorithms and system designs. Welcome to OpenDILab community! Scan the QR code and add us on Wechat:

Or you can contact us with slack or email ([email protected]).

License

DI-bioseq released under the Apache 2.0 license.

Citation

@misc{bioseq, title={{DI-bioseq: OpenDILab} Decision Intelligence platform for biological sequence prediction and searching}, author={DI-bioseq Contributors}, publisher = {GitHub}, howpublished = {\url{https://github.com/opendilab/DI-bioseq}}, year={2021}, }