Longitudinal Proteomics Biomarker Analysis and Trial Matching with LlamaIndex and Claude 4 Sonnet
Longitudinal blood proteomics studies reveal key insights into biological aging by tracking stable versus volatile protein expressivity across time. Quantifying individual trajectories in multi-year plasma proteomics enables early disease detection, dynamic risk scoring, and targeted clinical trial matching. However, synthesizing tabular longitudinal biomarker measurements alongside dense biomedical literature requires intelligent multi-modal indexing and advanced reasoning engines. In this challenge, you will construct a specialized Longitudinal Biomarker RAG and Trial Matching System using LlamaIndex and Claude 4 Sonnet. The system will ingest longitudinal protein assay datasets (such as normalized protein expression - NPX values for volatile markers like CXCL10, GDF15, and VEGFA), compute dynamic trajectory slope metrics, index relevant PubMed literature, and execute multi-stage retrieval workflows. Using LlamaIndex query engines and the extended context capabilities of Claude 4 Sonnet, your agent will generate comprehensive patient longitudinal reports, highlight volatile biomarker deviations, and automatically match patients with relevant active clinical trial protocols based on inclusion and exclusion criteria.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop a longitudinal proteomics biomarker analysis and clinical trial matching agent using LlamaIndex and Claude 4 Sonnet.
How work is evaluated
Evaluation framework validating protein trajectory trajectory classification accuracy and clinical trial matching precision.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
trajectory_classification_accuracy
Verifies correct categorization of stable vs volatile protein markers.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
trial_eligibility_precision
Checks precision of clinical trial matching against ground truth eligibility annotations.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
trial_matching_f1
F1 score for trial inclusion/exclusion evaluation precision. • target: 0.9 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
synthesis_hallucination_rate
Percentage of generated rationale statements unsupported by input trial docs. • target: 0.02 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Construct multi-modal knowledge indexes using LlamaIndex over longitudinal proteomic matrices and unstructured medical literature.
Implement trajectory feature extraction algorithms (e.g., rate of change, volatility index) on protein NPX series.
Integrate Claude 4 Sonnet for zero-shot clinical reasoning and multi-criteria clinical trial matching.
Build a clinical synthesis tool generating structured dynamic risk summaries with citation backing.
Reference links and supporting material
Synthetic longitudinal plasma proteomic assay dataset containing 500 patient records across 4 annual timepoints (2,925 proteins per sample), coupled with 100 clinical trial protocol documents.
Challenge setup
Evaluation framework validating protein trajectory trajectory classification accuracy and clinical trial matching precision.
Challenge input
JSON containing longitudinal NPX time-series matrix
LlamaIndex
Provides structured multi-modal indexing for combining tabular bio-assay metrics with unstructured trial docs.
Claude 4 Sonnet
Delivers superior long-context clinical reasoning and low hallucination in medical trial matching.
Scanpy / AnnData
Standard high-performance library for quantitative biological assay matrix manipulation.
Evaluated output
JSON with protein classifications, variance scores, and delta slopes
- Verifies correct categorization of stable vs volatile protein markers.
- Checks precision of clinical trial matching against ground truth eligibility annotations.
- F1 score for trial inclusion/exclusion evaluation precision. • target: 0.9 • range: 0-1
- Benchmark: ProteoTrialMatch-Bench
- Trial Matching F1 target: 0.9
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Llama Index · Optional
- Exa · Optional
- Claude 4 Sonnet · Required
Evaluation contract
- trajectory_classification_accuracy · Weight 1
- trial_eligibility_precision · Weight 1
- trial_matching_f1 · Weight 1
- synthesis_hallucination_rate · Weight 1
Metrics
- trial_matching_f1 · Target 0.9
- synthesis_hallucination_rate · Target 0.02
Benchmarks
- ProteoTrialMatch-Bench
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing longitudinal NPX time-series matrix
Output: JSON with protein classifications, variance scores, and delta slopes
Execution harness: Python
- Task 2
Input: JSON containing patient profile and candidate trial documents
Output: JSON list of eligible trials with match scores and criterion rationale
Execution harness: Python
Evaluation tools
- LlamaIndex · Data Indexing and Query Framework
- Claude 4 Sonnet · LLM Reasoning Engine
- Scanpy / AnnData · Proteomic Tabular Analysis
Recipe freeze marker
No recipe freeze time is recorded.