| { |
| "schema": "frontierlab.method-evaluation-draft.v1", |
| "recorded_date_taipei": "2026-10-28", |
| "source_handoff": "docs/runs/nim_method_handoff_20261028.json", |
| "status": "RESEARCHER_SELECTED_M2_READY_FOR_FORMAL_EVALUATION", |
| "selected_candidate_id": "M2", |
| "selection_rationale": "Researcher selected M2 because its structure-based computational design creates the clearest in-silico-to-in-vitro AI-for-Science demonstration: an AI agent can reason over a structural model, prioritize dual-pH histidine-scanning variants, and hand candidates to experimental Fab, cell, full-antibody and functional validation. This produces a stronger end-to-end FrontierLab narrative than M1, whose core search loop depends more heavily on physical library generation and FACS screening. M1 is retained as a comparator because it can reveal histidine-independent pH-switching solutions and therefore exposes the selected method's histidine-centered bias. M2's assay limitations are treated as a validation requirement rather than ignored: downstream planning should include an orthogonal biophysical check such as SPR on monomeric Fab where feasible to distinguish intrinsic affinity from avidity/stoichiometry effects.", |
| "candidates": [ |
| { |
| "candidate_id": "M1", |
| "name": "Directed evolution with yeast surface display and alternating-pH selection", |
| "evidence_ids": [ |
| "PMC4314675" |
| ], |
| "scope": "Use a displayed Fcab library with positive selection for Her2 binding at pH 7.4 and negative selection/non-binding at pH 6.0, followed by clone-level pH-dependent binding characterization.", |
| "comparison": { |
| "expected_value": "Directly explores sequence space and can discover pH-sensitive variants without assuming histidine substitutions are required.", |
| "resources": "Yeast display library, flow cytometric sorting, recombinant Her2, soluble-expression workflow and downstream binding/stability assays.", |
| "validation": "Compare binding at pH 6.0 versus 7.4, then verify soluble-protein/cell binding and stability with independent repeats.", |
| "limitations": "Empirical and resource-intensive; findings may depend strongly on library design, target, antibody format and selection thresholds.", |
| "novelty": "Useful discovery-oriented comparator because it can reveal non-obvious pH-switching sequence solutions." |
| }, |
| "reproducibility_checks": { |
| "data": "UNKNOWN_PENDING_FULL_TEXT_EXTRACTION", |
| "code": "NOT_APPLICABLE_OR_NOT_YET_IDENTIFIED", |
| "environment": "UNKNOWN_PENDING_FULL_TEXT_EXTRACTION", |
| "parameters": "PARTIALLY_DOCUMENTED_IN_SOURCE", |
| "protocol": "DOCUMENTED_IN_SOURCE_NOT_REPRODUCED", |
| "metrics": "DOCUMENTED_IN_SOURCE_NOT_REPRODUCED" |
| }, |
| "executed_in_frontierlab": false |
| }, |
| { |
| "candidate_id": "M2", |
| "name": "Structure-based dual-pH histidine-scanning computational design", |
| "evidence_ids": [ |
| "PMC6927761" |
| ], |
| "scope": "Use a structure-based antibody design platform with dual-pH histidine-scanning mutagenesis to propose variants, then test Fab and full-size antibody binding across acidic and physiological pH.", |
| "comparison": { |
| "expected_value": "Provides a targeted, mechanism-guided design route that can reduce the experimental search space and optimize acidic-pH selectivity.", |
| "resources": "Antibody-antigen structural model, computational design platform, recombinant expression, binding assays, cell-based validation and functional spheroid assays.", |
| "validation": "Measure recombinant and cell-surface binding across pH, compare parental controls, inspect Her2-density effects and confirm functional activity under acidic versus physiological conditions.", |
| "limitations": "Depends on structural information and histidine-centered design assumptions; assay stoichiometry and avidity can affect apparent affinity estimates.", |
| "novelty": "Useful design-oriented comparator because it prospectively links structural modeling to pH-selective antibody engineering." |
| }, |
| "reproducibility_checks": { |
| "data": "UNKNOWN_PENDING_FULL_TEXT_EXTRACTION", |
| "code": "UNKNOWN_PLATFORM_AVAILABILITY", |
| "environment": "UNKNOWN_PENDING_FULL_TEXT_EXTRACTION", |
| "parameters": "PARTIALLY_DOCUMENTED_IN_SOURCE", |
| "protocol": "DOCUMENTED_IN_SOURCE_NOT_REPRODUCED", |
| "metrics": "DOCUMENTED_IN_SOURCE_NOT_REPRODUCED" |
| }, |
| "executed_in_frontierlab": false |
| } |
| ], |
| "automatic_selection": false, |
| "automatic_planning": false, |
| "scientific_execution_verified": false, |
| "note": "Stage 4 researcher selection recorded. M2 is selected for the FrontierLab demo path; M1 remains a comparator. Planning remains unapproved and blocked.", |
| "researcher_decision": { |
| "reviewer": "Jaycee871", |
| "recorded_date_taipei": "2026-10-28", |
| "decision": "SELECT_M2", |
| "automatic_selection": false, |
| "planning_approved": false, |
| "key_reasoning": [ |
| "M2 provides the strongest in-silico-to-in-vitro AI-for-Science demonstration space.", |
| "M2 supports a clearer end-to-end computational-design-to-biological-validation storyline.", |
| "M1 remains useful as a comparator but depends more heavily on physical library generation and FACS screening.", |
| "Avidity/stoichiometry limitations should become explicit downstream validation checks rather than be hidden." |
| ], |
| "proposed_downstream_check": "Where feasible, use an orthogonal assay such as SPR on monomeric Fab to distinguish intrinsic affinity from avidity effects." |
| } |
| } |
