Skip to main content

Assess Splice Isoform Functionality

TRIFID is a machine learning-based tool that predicts the functional relevance of splice isoforms using Random Forest models and SHAP interpretability.

Get StartedAPI Reference
# Install TRIFID
pip install git+https://github.com/fpozoc/trifid.git
# Make predictions
python -m trifid.models.predict

Quick Start

Get up and running with TRIFID in minutes

1

Install TRIFID

Install TRIFID directly from GitHub using pip:
pip install git+https://github.com/fpozoc/trifid.git
For development, clone the repository and install with conda/mamba:
git clone https://github.com/fpozoc/trifid.git
cd trifid
mamba env create -f environment.yml
conda activate trifid
2

Load Pre-computed Predictions

Access predictions for human and other species from available datasets:
import pandas as pd

# Load TRIFID predictions for GENCODE 27
predictions = pd.read_csv(
    'data/genomes/GRCh38/g27/trifid_predictions.tsv.gz',
    compression='gzip',
    sep='\t'
)

# Filter by gene name
gene_predictions = predictions[predictions['gene_name'] == 'FGFR1']
print(gene_predictions[['transcript_id', 'trifid_score', 'appris']])
Pre-computed predictions are available for GENCODE, RefSeq, and Ensembl annotations across multiple species including human, mouse, rat, and more.
3

Train a Custom Model

Train your own TRIFID model on custom genome annotations:
# Prepare your dataset
python -m trifid.data.make_dataset

# Train the model
python -m trifid.models.train

# Generate predictions
python -m trifid.models.predict
4

Interpret Results with SHAP

Use SHAP values to understand which features drive predictions:
from trifid.models.interpret import explain_prediction

# Load SHAP explanations for a specific isoform
explain_prediction(df_shap, model, features, 'ENST00000356207')
TRIFID’s interpretability features help identify which predictive features (splice junction coverage, domain conservation, etc.) contribute most to each isoform’s functional score.

Key Features

Powerful capabilities for splice isoform functional assessment

Random Forest Classifier

Accurate predictions using ensemble machine learning trained on large-scale proteomics data

SHAP Interpretability

Understand feature importance and model decisions with SHAP-based explanations

QSplice Module

Quantify splice junction coverage from RNA-seq data to assess isoform expression

Pfam Effects Analysis

Evaluate the impact of alternative splicing on protein domain integrity

Multi-Species Support

Pre-computed predictions for human, mouse, rat, and other vertebrate and invertebrate species

Multiple Genome Builds

Support for GENCODE, RefSeq, and Ensembl annotations across different genome assemblies

Explore by Use Case

Find the documentation you need for your workflow

Data Preparation

Learn how to prepare genome annotations, RNA-seq data, and feature datasets for TRIFID

Read guide

Training Models

Train custom Random Forest models on your own datasets with hyperparameter optimization

Read guide

Making Predictions

Apply trained models to predict functional relevance scores for any set of isoforms

Read guide

Interpreting Results

Use SHAP values and visualization tools to understand what drives each prediction

Read guide

Resources

Additional resources to help you succeed with TRIFID

Case Studies

Explore real-world examples like FGFR1, C1orf112, and NIPAL3

Research Paper

Read the published manuscript in NAR Genomics and Bioinformatics

GitHub Repository

Access the source code, notebooks, and contribute to the project

Ready to Get Started?

Install TRIFID and start predicting functional relevance of splice isoforms in your research

Installation GuideView Tutorial