Business Context

Quantera Bio runs high-sensitivity Simoa immunoassays to detect low-abundance protein biomarkers in blood and cerebrospinal fluid. The assay team wants a model that classifies whether a sample is biomarker-positive and estimates concentration quality from instrument and assay-derived features, reducing manual review while preserving scientific trust.

Dataset

You are given historical assay runs collected from a Simoa platform. Simoa (Single Molecule Array) works by isolating immunocomplexes on beads into femtoliter-sized wells so fluorescence can be measured at near single-molecule resolution. Each row represents one processed sample from a completed run.

• Size: 48K samples across 14 months, 46 features
• Target: Binary classification — biomarker-positive (1) vs negative (0)
• Class balance: Moderately imbalanced — 18% positive, 82% negative
• Missing data: 6% missing in sample metadata, 3% missing in QC fields, occasional unseen reagent lots in test data

Success Criteria

A good solution should achieve strong recall on positive samples without overwhelming the lab with false positives. Target ROC-AUC > 0.88, PR-AUC > 0.60, and recall >= 0.85 at precision >= 0.55.

Constraints

• Predictions must be explainable to assay scientists and QA reviewers
• Batch inference should score 5K samples in under 2 minutes
• Model retraining is allowed monthly, not per run
• The solution must be robust to lot changes and mild feature drift

Deliverables

1. Build a classification pipeline for biomarker positivity.
2. Explain how Simoa principles inform feature selection and model choice.
3. Describe preprocessing for mixed data types, missing values, and lot variability.
4. Evaluate the model with appropriate metrics and threshold selection.
5. Propose a deployment and monitoring plan for monthly retraining and lab QA review.