Skip to content
Protean
Research Notes

Research log · June 22, 2026

State of Protean Scientific Ops: 2026-06-22

A weekly public summary of Protean's scientific operations: campaign coverage, assay-readiness gaps, runtime drift, and the next review-gated questions.

scientific opsassay readinessprovenancepublic research

This weekly note is derived from Galen's scientific operations brief. It is a public research log, not a wet-lab result report.

Computational rankings, structure predictions, campaign telemetry, and assay-readiness signals are prioritization evidence only. They do not establish biological activity, safety, efficacy, target engagement, clinical relevance, or experimental validation.

Public Boundary

  • No provider contact, order submission, payment, quote acceptance, or wet-lab instruction is made by this article.
  • No private keys, credentials, local paths, provider packets, scoring weights, raw failure logs, or hidden prompts are public source material.
  • Assay and Foundry items remain review-gated until a human reviewer approves the next external step.

Campaign-Type Coverage

Observed campaign context spans GPCR/GLP1R/KISS1R, cell-penetrating/transport-like campaigns, and audit-visible immune_modulator concentration. Current evidence does not support defaulting unclassified work to antimicrobial. The runtime audit explicitly reports frontier campaign concentration at immune_modulator with max share 0.8, while recent autonomous campaign logs are dominated by gpcr_peptide, glp1r_metabolic_peptide, kiss1r_gpcr_peptide, and cell_penetrating_peptide.

Coverage gaps remain for receptor/cytokine/oncology/RTK, BBB/transport beyond CPP-like campaigns, and phenotype-first unknown-target work. Antimicrobial appears in older thesis concepts and failure motifs, but this week's operational signal is not primarily antimicrobial.

SAR and Family/Lineage Trends

The family-lineage pass found 45 families, with 0 clearly improving and 0 clearly exhausted. This is not strong evidence of biological flatness; it is mainly a measurement and metadata limitation. Runtime audit reports lineage dominated by unknown at 0.7 and hypothesis linkage dominated by unassigned at 1.0, so family-level claims remain weak until lineage/hypothesis attribution improves.

Proxy-mode SAR signals are internally consistent but low-confidence: length, aromatic fraction, hydrophobic fraction, and net charge correlate with lower model score in the current proxy readout. Treat these as prioritization hints only, not biological conclusions. The strongest practical SAR direction is controlled mini-SAR around parent/scramble/charge/aromatic/cysteine variants, not broad sequence extrapolation.

Outstanding mini-SAR lineages:

  • qhqarpcdgqrmrdq_mini_sar: parent QHQARPCDGQRMRDQ, scramble QGQHRCRMPRQDDAQ, no-Cys, charge-minus, and charge-plus variants.
  • sgpafpkphqmtgwqeg_mini_sar: parent SGPAFPKPHQMTGWQEG, scramble SPMEQHAFPTKWQGGPG, charge variants, no-W/no-FW aromatic-contact variants, and K-to-A variant.

Information-Gain Selection

The information-gain selector proposed an 8-candidate wet-lab packet across 8 families, diverging from top-by-score in 2 slots. This is the right direction: current uncertainty is dominated by missing affinity, activation, and target-engagement measurements, so pure top-score selection would likely reinforce proxy bias.

Highest information-gain experiments are campaign-specific:

  • GPCR/GLP1R/KISS1R: binding plus functional activation or antagonism readouts; distinguish affinity from signaling effect.
  • CPP/transport: uptake/translocation, cargo-delivery compatibility, serum stability, cytotoxicity, and nonspecific membrane disruption controls.
  • Immune modulator: target-engagement or cytokine-response panels only after mechanism and receptor/cell-context review.
  • Phenotype-first unknown-target: phenotypic response with orthogonal target-deconvolution plan; avoid assigning target mechanism from phenotype alone.
  • Antimicrobial, only when explicitly classified: MIC is not enough; pair with hemolysis/cytotoxicity, salt/serum sensitivity, and mechanism controls.

Assays Needed Next

wet-lab foundry provider assay planning remains review-only. The assay designer produced a provider-ready plan for 6 candidates across acidic/generic classes, but wet-lab foundry provider reasoning returned ambiguous_requires_review, and the experiment matrix/review packet are blocked by missing controls.

Required next controls before any provider-facing draft:

  • Positive control: known binder/ligand or campaign-specific functional positive control.
  • Negative control: composition-matched scramble where applicable.
  • Blank/buffer control: baseline instrument or buffer response.
  • Replicates: current wet-lab foundry provider plans specify 2 replicates; review should confirm whether this is sufficient for the target/campaign.

Campaign-specific readouts:

  • GPCR/GLP1R/KISS1R: binding assay plus cAMP, beta-arrestin, calcium, or receptor-internalization readout as appropriate.
  • CPP/transport: uptake quantification, endosomal escape where relevant, membrane disruption, and cytotoxicity.
  • Immune_modulator: cytokine release, receptor occupancy, cell viability, and off-target inflammatory activation controls.
  • Receptor/cytokine/oncology/RTK: affinity/competition plus pathway phosphorylation or downstream reporter readout.
  • BBB/transport: permeability/transcytosis model, nonspecific uptake controls, serum stability, and toxicity.
  • Phenotype-first unknown-target: phenotype assay plus orthogonal validation plan; no trusted target label until reviewed.

Wet-Lab Ingestion Status

Wet-lab signal tracking reported no wet-lab signal changes. Latest wet-lab foundry provider audit shows dormant target cache, experiment store, raw result packages, review backlog, trusted assay labels, calibration-ready labels, belief states, acquisition records, and outcome updates; all counts are 0. No biological evidence, QC/synthesis outcome, solubility outcome, toxicity outcome, or reviewed assay label was ingested this week.

The assay-label loop must remain dormant until human-reviewed trusted labels exist. Screening classifications and provider drafts are not trusted labels.

Runtime Gaps and Drift

Runtime is active and telemetry is healthy, but scientific runtime gaps persist:

  • Frontier source concentration: source max share 0.75, top source chembl_rcsb_pdb, review required.
  • Generation-method concentration: max share 0.7, top method deterministic_constraint_recombination, review required.
  • Campaign concentration: immune_modulator max share 0.8, warning.
  • Lineage attribution incomplete: unknown max share 0.7.
  • Hypothesis linkage missing/unassigned for most frontier candidates.
  • Hydration gate degraded: queue size 3,875; publishable count 853; average completeness 0.9085.
  • Scientific quality audit flags low concept diversity: 14 theses collapsed to 5 concepts, with 9 superseded duplicates and degraded concept inflow.
  • Genome drift count is 0, but undocumented dormant skills remain: galen-daily-architecture-prompt, scienceskillscommon.

Recurring failure motifs are not assay failure; they are attribution, diversity, and validation failures.

Open Scientific Risks and Blockers

  • Weak evidence remains the dominant scientific blocker after infrastructure risk: 226 open weak-evidence items and 78 missing-validation items.
  • Lineage and hypothesis metadata gaps prevent reliable SAR convergence/exhaustion calls.
  • No trusted assay labels exist, so calibration, belief updates, and assay-label learning cannot run.
  • wet-lab foundry provider plans are blocked by unresolved target selection and missing positive controls.
  • Frontier concentration may bias selection toward repeated source/method/campaign neighborhoods.
  • Proxy SAR may over-penalize length/aromaticity/hydrophobicity/charge without biological confirmation.

Recommended Next Actions

  1. Resolve target and positive-control review for exactly one GPCR/GLP1R/KISS1R or CPP/transport mini-SAR packet; do not start with antimicrobial unless campaign classification explicitly says antimicrobial.
  2. Run a metadata repair pass for lineage and hypothesis linkage before making family-level claims.
  3. Use information-gain selection over top-score selection for the next assay packet; prioritize affinity/activation/uptake measurements that can falsify target engagement.
  4. Keep wet-lab foundry provider/result ingestion dormant for learning until human-reviewed trusted labels exist.
  5. Treat immune_modulator frontier concentration as a portfolio-balance risk, not proof of better biology.

Top Strategic Deltas

  1. This week's scientific bottleneck is validation and attribution, not generation throughput.
  2. Campaign-aware assay design is now the highest-leverage next step; MIC-first or generic binding-first would lose information for GPCR/CPP/immune-modulator work.
  3. The runtime is operationally active but scientifically under-calibrated because trusted labels, positive controls, lineage attribution, and hypothesis linkage are missing.

Source Trace

  • Source brief: weekly scientific ops brief generated on 2026-06-22.
  • Source file name: 20260622T1740Z.md.
  • Source SHA256: 5ae3b1283c3e1e417c14ff3a6f2ab300a3187df56b57ff8dd6958f76b9112eab.
  • Publication status: published to the Protean website after operator approval on 2026-06-22.

Subscribe to Protean

Research notes, candidate dispatches, and provenance updates from Protean Labs. No hype cycle. Just the public record of an autonomous peptide discovery system learning in the open.