Local Similarity Analysis (LSA) of Time Series Data and Applications to Metagenomics
summary
Method detects short, time-local and time-delayed associations between OTU abundance series that Pearson correlation misses, maximizing a local similarity score on rank-normalized monthly OTU abundance and allowing short lags; it derives theoretical P-values from Feller’s theorem (conservative tails) and uses a hybrid simulation-only refinement for very small P-values; tested on 18+ years of SPOT monthly OTU data (16S tag clusters) and wins comparative benchmarks (Rob Knight / ISME) against other network methods.
AI-written summary · about
chapters (17)
- 0:00Intro
- 0:28Protein Interactions and Applications
- 9:49USC San Pedro Ocean Time-series (SPOT) and Microbial Observatory (MO) (PI: Jed Fuhrman)
- 12:09Abundant Metagenomic Time Series Data
- 14:45Biological Questions
- 17:04OTU Association Network
- 20:05Local Association
- 20:45Time-delayed Association
- 21:56Data Normalization
- 22:47Local Similarity Analysis (LSA)
- 23:40Dynamic Programming Algorithm for Calculating LS
- 23:56Extended Local Similarity Analysis (eLSA): Dealing with Replicates
- 24:49Bootstrap Confidence Interval for LS Score
- 26:31Range of Partial Sum of ID Mean Zero Variables (Feller 1951)
- 28:01Applications to LS Statistical Significance
- 28:38Theoretical vs Simulated p-values: D=0
- 30:10Suggested Methods
related talks
36mExtracting Relevant Metrics with Spectral Clustering
40mData Deduplication using Locality Sensitive Hashing
34mTopological Data Analysis
49mSpectral Approaches to Nearest Neighbor Search
31mLearning Latent Events from Network Message Logs
26mHow to Apply the Machine Learning Appropriate Way for Your Security Operation
49mIntroduction to Biological Network Analysis IV: Network Alignment and Querying
10mThrough the Lens of LLMs: Unveiling Differential Privacy Challenges