by-experiment-results

You can calibrate in-silico predictions by ingesting wet-lab readouts to form a canonical dataset, then computing calibration metrics like AUC and Mann-Whitney U to generate a calibration report.

The diagnosis applies Mann-Whitney U tests with BH correction and computes AUC values to evaluate in-silico predictions against ground-truth lab outcomes.

Yes, the calibration process relies on pandas, numpy, scipy, and scikit-learn to normalize lab batches, join in-silico features, and compute statistical calibration metrics.

You can push validated or contradicted wet-lab findings to a knowledge graph by running a dedicated update script that processes the normalized lab results and enriched dataset.

The best way is to join per-design in-silico features with previous lab readouts to compute calibration metrics, producing an enriched dataset for the next campaign optimizer.

You need a lab readout file that the ingestion script can normalize and join with in-silico features to produce the canonical calibration dataset required for diagnosis.