A systems approach for analysis of high content screening assay data with topic modeling

Area of Science:

• Toxicology and Pharmacology
• Systems Biology
• Computational Biology

Background:

• High Content Screening (HCS) enables simultaneous measurement of multiple cellular endpoints for toxicity assessment.
• Understanding systems biology requires analyzing probabilistic relationships between these endpoints under stress.
• Extracting hidden knowledge from complex HCS data presents a significant challenge.

Purpose of the Study:

• To apply Latent Dirichlet Allocation (LDA) for analyzing in vitro HCS cellular endpoints.
• To correlate HCS findings with in vivo histopathological observations, specifically drug-induced necrosis.
• To identify diagnostic topics linking specific cellular responses to observed toxicological outcomes.

Main Methods:

• Converted continuous HCS measurements into document-term format (word frequency) for LDA analysis.
• Generated 488 documents (drug-time points) with 10 endpoints treated as words.
• Applied LDA to extract three topics and identify diagnostic patterns for 45 common drugs.

Main Results:

• Assay endpoints clustered into topics that corresponded with observed in vivo histopathology (necrosis).
• Early necrosis (6 hours) was linked to severe damage markers (Steatosis, DNA Fragmentation, Mitochondrial Potential, Lysosome Mass).
• Later necrosis (24 hours) was associated with DNA Damage and Apoptosis, suggesting pathway interplay; absence of necrosis correlated with Cell Loss and Nuclear Size, indicating regeneration.

Conclusions:

• Topic modeling using LDA effectively interprets relationships between in vitro HCS endpoints and in vivo necrosis findings.
• This computational approach enhances the understanding of systems biology and cellular responses to chemical treatments.
• LDA offers a powerful method for integrating multi-endpoint assay data for toxicological assessments.