Skip to content

Base Layer Architecture

Purpose: Foundation for splice site prediction with extensible base model support
Component: splice_engine/base_layer/
Status: 🚧 Design complete, refactoring in progress

Overview

The base layer provides an abstract interface for splice site prediction models, allowing: - Multiple base models (SpliceAI, OpenSpliceAI, future models) - Standardized prediction workflow - Consistent data formats - Coordinate system handling

Current Status

This is an active development area. Key components are implemented; workflow orchestration and chunking are in progress.

Key Concepts

1. Abstract Base Model Interface

All base models implement a common interface:

class BaseModel(ABC):
    """Abstract interface for base prediction models."""

    @abstractmethod
    def predict(
        self,
        sequence: str,
        gene_annotations: pd.DataFrame
    ) -> Dict[str, np.ndarray]:
        """Predict splice sites for a sequence."""
        pass

2. Coordinate System Conventions

CRITICAL: Base models have different coordinate conventions: - SpliceAI: 0-based, predictions at position - OpenSpliceAI: 0-based, predictions offset by +1

See Coordinate Adjustment for the full explanation, known offsets, and implementation guide.

3. Prediction Pipeline

Standardized workflow: 1. Load genome and annotations 2. Prepare analysis sequences 3. Run base model predictions 4. Adjust coordinates (model-specific) 5. Detect peaks 6. Save artifacts

Architecture Components

1. Base Model Abstraction

# Concrete implementations
class SpliceAIModel(BaseModel):
    """SpliceAI base model wrapper."""
    coordinate_offset = 0

class OpenSpliceAIModel(BaseModel):
    """OpenSpliceAI base model wrapper."""
    coordinate_offset = 1

2. Data Preparation

# Prepare analysis sequences
from splice_engine.base_layer.data_prep import prepare_analysis_sequences

sequences = prepare_analysis_sequences(
    genes=genes,
    genome=genome,
    context_length=10000
)

3. Prediction Workflow

# Run predictions
from splice_engine.base_layer.workflow import run_base_predictions

results = run_base_predictions(
    base_model="spliceai",
    chromosome="chr1",
    genes=genes,
    genome=genome
)

Refactoring Progress

Component Status Notes
Base model interface ✅ Designed Abstract class defined
SpliceAI wrapper ⚠️ In progress Coordinate adjustment needed
OpenSpliceAI wrapper ⚠️ In progress Coordinate adjustment needed
Data preparation ⚠️ Partial Core functions ported
Prediction workflow 📝 Planned Not yet implemented
Coordinate adjustment ✅ Core done Integration pending

Next Steps

  1. ✅ Complete coordinate adjustment integration
  2. 🚧 Refactor prediction workflow
  3. 📝 Implement chunking and checkpointing
  4. 📝 Add comprehensive testing

Status: 🚧 Active development area
Last Updated: February 15, 2026
Next Review: After Phase 1.5 completion