lncRNA-Py Documentation

lncRNA-Py is a development package for applying machine learning and deep learning to the problem of lncRNA classification, i.e. predicting whether a novel RNA transcript is coding (mRNA) or long non-coding (lncRNA).

The main functionalities of the package are:

• lncRNA-BERT: A Nucleotide Language Model (NLM) pre-trained on human mRNA and lncRNA data, achieving state-of-the-art performance when fine-tuned for lncRNA classification.

• Convolutional Sequence Encoding: A novel encoding technique that enables a more efficient and effective accomodation of long nucleotide sequences in comparison to K-mer Tokenization and Byte Pair Encoding. Alternative encoders (NUC, K-mer, BPE) are also implemented.

• Library of predictor features: Re-implementations of a large collection of coding potential predictors (e.g. ORF length, pI, Fickett score, k-mer frequencies) used by over 40 existing lncRNA classifiers.

Usage

There are two ways in which lncRNA-Py can be used: 1) running pre-written scripts; or 2) via the object-oriented API. The scripts are dedicated to (pre-)training NLMs, the API provides access to the extensive library of re-implemented features.

We provide two example notebooks, one for using our pre-written scripts, and a second one that introduces the API.

• Installation
  • Required Dependencies
  • Optional Dependencies
• Running Scripts
  • Classify
  • Train
  • Pre-Train
  • Generate Embeddings
  • Byte Pair Encoding
• API Documentation
  • lncrnapy.data
  • lncrnapy.features
  • lncrnapy.modules
  • lncrnapy.train
  • lncrnapy.evaluate
  • lncrnapy.utils
  • lncrnapy.algorithms
  • lncrnapy.selection
• Experiments
  • Main Experiments
  • Hyperparameter Tuning

Publication: lncRNA-BERT

LncRNA-Py accompanies the paper: LncRNA-BERT: An RNA Language Model for Classifying Coding and Long Non-Coding RNA. Scripts and descriptions of additional experiments that relate to this study are provided in the experiments folder.

License

LncRNA-Py is available under the MIT license.