OpenSpliceAI

OpenSpliceAI implements an open-source, modular reimplementation of SpliceAI in PyTorch for efficient splice site prediction and retraining across various species.

Key Features:

• Modular Architecture: The framework provides tools for creating training datasets, training models, predicting splice sites, and assessing the impact of genetic variants.
• Training Flexibility: Supports both training from scratch and transfer learning, enabling seamless retraining on species-specific datasets and mitigating human-centric biases.
• Computational Efficiency: Achieves faster processing speeds and lower memory usage than the original SpliceAI code, allowing large-scale analyses of extensive genomic regions on a single GPU.
• Ecosystem Integration: Flexible architecture enables easier integration with established machine learning ecosystems, simplifying the development of custom splicing models for different species and applications.
• Model Concordance: Output is highly concordant with SpliceAI, with in silico mutagenesis analyses confirming both models rely on similar sequence features and calibration experiments demonstrating similar score probability estimates.

Scientific Applications:

• Cross-Species Splice Site Prediction: Enables retraining on species-specific datasets for splice site prediction in nonhuman species.
• Variant Impact Assessment: Assesses the impact of genetic variants on splicing.
• Large-Scale Genomic Analysis: Supports analysis of extensive genomic regions through efficient processing on a single GPU.
• Custom Splicing Model Development: Facilitates development of custom splicing models for different species and applications through integration with machine learning ecosystems.

Methodology:

OpenSpliceAI is implemented in PyTorch as a reimplementation of the SpliceAI deep learning system, using in silico mutagenesis analyses and calibration experiments to validate concordance with the original SpliceAI model.