fastLSA

fastLSA identifies local and time-lagged associations among co-varying time series and evaluates their statistical significance using an asymptotic p-value bound to enable scalable analysis of large datasets.

Key Features:

• Local Similarity Analysis (LSA) statistic: Computes LSA to detect local and lagged relationships between time series.
• Asymptotic p-value bound (no normality assumption): Derives an asymptotic upper bound for p-value calculations without assuming normality, replacing computationally intensive permutation tests.
• Computational complexity reduction: Reduces significance-testing complexity from O(pm²n) to O(m²n), where p is permutations, m is number of series, and n is series length.
• Multi-core C implementation: Implemented in C with threading to leverage multi-core processors for high-throughput computation.
• Network visualization support: Produces network outputs compatible with Cytoscape for force-directed graph visualization and mapping of metadata to explore correlated cliques.

Scientific Applications:

• Public health surveillance: Identify temporal associations in epidemiological and surveillance time series.
• Microbial ecology: Uncover time-lagged relationships and temporal dynamics among microbial taxa.
• Social media analytics: Detect lagged correlations and temporal patterns in social media activity data.
• Temporal network discovery: Reveal correlated cliques and network structures derived from large-scale time series.

Methodology:

Computes the Local Similarity Analysis statistic and derives an asymptotic upper bound for p-value calculations to obviate permutation tests and avoid normality assumptions; achieves reduced complexity from O(pm²n) to O(m²n) and is implemented in C with threading for multi-core execution.