Tomtom-lite

Tomtom-lite re-implements the Tomtom algorithm to accelerate motif similarity scoring for large-scale genomic analysis of continuous representations such as position-weight matrices (PWMs).

Key Features:

• Numba acceleration: The re-implementation uses numba acceleration as a Python function.
• Multi-threading: The tool is natively multi-threaded.
• Cache optimization: It avoids cache misses and more efficiently caches intermediate values.
• Computational approximations: It uses approximations at compute bottlenecks.
• Large-scale speedup: It achieves over a 1000-fold speedup compared with the original tool on reasonable tasks.

Scientific Applications:

• PWM similarity scoring: It identifies pairs of position-weight matrices (PWMs) whose similarity is statistically significant.
• Contribution-weight matrix analysis: It calculates similarities for attribution-based contribution-weight matrices.
• Real-time motif comparison: It enables real-time motif similarity scoring that was previously inefficient.
• Large-scale genomic analysis: It supports large-scale genomic analysis involving continuous representations.

Methodology:

Tomtom-lite re-implements the original Tomtom method using numba acceleration, native multi-threading, cache miss avoidance, efficient caching of intermediate values, and approximations at compute bottlenecks.