RELSA Severity Assessment & Humane Endpoint Forecasting

What it does

• RELSA scoring: merges multiple welfare readouts (weight loss, temperature, clinical scores, inflammatory biomarkers, activity, heart rate, burrowing) into one relative severity score per animal per time point (0 = baseline, 1 = reference set maximum).
• Humane endpoint forecasting (foRcast): fits an ARIMA model to an individual animal's RELSA trajectory, predicts the next score with a 95% prediction interval, and evaluates it with RMSE, PICP and MPIW.
• Severity zones: uses kernel density estimation to locate density minima as candidate attention/danger thresholds, with a required bandwidth sensitivity sweep.
• Reporting discipline: documents the four decisions that drive results (directionality, baseline normalization, reference set, zero-baseline score mapping), nine common pitfalls, explicit regulatory boundaries and a reproducibility checklist.

Who it's for

• Researchers, veterinarians and animal facility staff performing severity assessment.
• Anyone writing the severity section of a 3Rs/refinement analysis or an EU Directive 2010/63/EU application.
• Teams wanting to compare multivariate welfare data (telemetry, clinical scoring) on one common scale.

Examples

1. Score a cohort: run relsa_score.py on assets/example_cohort.csv (6 mice, 9 days) with weight, temperature, a 0–8 clinical score and IL-6, select the endpoint group as reference, save it to JSON and export per-day RELSA plus per-variable weights.
2. Flag at-risk animals early: train up to the time point before euthanasia and predict the endpoint score (e.g. 0.93 [0.67–1.19] vs actual 1.00); for live monitoring use --mode rolling one-step-ahead forecasts.
3. Define zones: run kde_thresholds.py to find a threshold at 0.703 and report it together with a bandwidth sweep, since a 10% larger bandwidth can remove the minima entirely.

Install the skill I found on Claude Skill Mart.
Copy the skills/relsa-severity-assessment folder from the GitHub repo K-Dense-AI/scientific-agent-skills into my ~/.claude/skills/relsa-severity-assessment/.
When it's done, tell me in one line what this skill can do.