Peptiter · DiscoveryLab

Condition-first peptide discovery without brute-force sequence search.

DiscoveryLab turns therapeutic hypotheses into receptor-aware peptide families, ranks them through in-silico assessment, visualizes peptide–receptor fit in 3D, and hands validated batches to wet-lab partners through LabSpace.

Search reduction
10¹³ → 10²
per program
Receptor families
GPCR · NHR · CK
examples
Wet-lab loop
LabSpace
feedback-driven
FIG-01 · DiscoveryLab session viewcondition → families → fit
discoverylab · session #DL-2841 · obesity / metabolic-dysfunction
v1.4.2
01 · Condition
Obesity / metabolic dysfunction
ICD-10: E66 · MeSH: D009765
02 · Mapped pathways
  • Incretin signaling0.92
  • Melanocortin axis0.81
  • Leptin / JAK-STAT0.74
  • Adipocyte lipolysis0.61
03 · Receptor set
GLP-1RGIPRGCGRMC4RGHRH-R
Search-space reduction
Sequence universe (10-mer)
1.02e13
Receptor-conditioned scaffolds
8.4e6
BioScout source-system motifs
142,300
Family neighborhoods
12,448
Passed in-silico gates
74
3D · GLP-1R · candidate DL-PEP-0421
His-7 · key contactC-term · stabilized
RealityKit · receptor 7TM · ribbon
ΔG est. −9.2 kcal/mol
Candidate families
  • DL-FAM-0140.88
    GLP-1R agonist scaffold · n=32
  • DL-FAM-0190.81
    Dual GLP-1R / GIPR · n=24
  • DL-FAM-0220.74
    MC4R selective · n=18
In-silico lab score · DL-PEP-0421
  • Receptor fit0.88
  • Stability (predicted)0.71
  • Solubility0.83
  • Aggregation risk0.18
  • Synthesis feasibility0.79
● connected · LabSpace partner: Helix Bio Lab #L-228candidates evaluated 12,448 / 12,448 · passed gates 74
The problem

Peptide design fails when the search space is unconstrained.

The number of possible peptides grows exponentially with sequence length. Even short peptides yield combinatorial spaces that no generation strategy can meaningfully cover by sampling alone.

10-mer
20¹⁰
≈ 1.024 × 10¹³
natural amino acids
14-mer
20¹⁴
≈ 1.64 × 10¹⁸
natural amino acids
14-mer + ncAA
(20+k)¹⁴
intractable
with non-natural residues

DiscoveryLab treats peptide design as a search-space optimization problem, not a text-generation problem. Generation, ranking, and rejection happen at every stage — under biological, structural, and synthesis constraints.

Search-space optimization

How DiscoveryLab avoids exponential explosion.

Eight stages, each constraining, ranking, or rejecting candidates. The full peptide universe is never enumerated — only biologically plausible neighborhoods are searched.

FIG-02 · Cumulative search-space reduction (representative program)log scale
Stage 0 · raw 10-mer universe
1.02e13
Stage 2 · receptor-conditioned
8.4e6
Stage 3 · BioScout source motifs
1.42e5
Stage 4 · scaffold-constrained
4.8e4
Stage 5 · family neighborhoods
1.24e4
Stage 6 · GP-evolved survivors
2,180
Stage 7 · passed in-silico gates
74
Stage 8 · wet-lab batch
12
STAGE 01

Condition-first framing

Researchers begin with a biological condition or desired phenotype. The system maps it to pathways, receptor families, known ligand classes, contraindication concerns, relevant assays, and clinical/preclinical evidence.

obesityrheumatoid arthritismetabolic dysfunctioninflammationtissue repairsarcopeniaendocrine signalingimmune modulation
STAGE 02

Target and receptor narrowing

Instead of searching all possible peptides, DiscoveryLab selects a constrained biological target set — limited to evidence-supported receptors and pathways for the condition.

GLP-1RGIPRGCGRMC4RGHRH-RIL pathwaysTNF-relatedGPCRsnuclear hormone

Examples only. Coverage depends on available evidence per condition.

STAGE 03

BioScout mimicry

Comparative BioScout identifies evolved source systems that already solve related biological problems and produces evidence-backed source-system search plans — not blind candidate generation.

venom peptidesgut–brain signalinghost defenseendocrine peptideswound repairimmune-modulatingcyclic / constrained
STAGE 04

Motif and scaffold constraints

The system extracts or applies pharmacophore hints, conserved motifs, receptor-binding residues, charge and hydrophobicity profiles, cyclization opportunities, secondary-structure propensity, protease resistance, and synthesis feasibility.

motifspharmacophoresbinding residuescharge mapcyclizationhelicityβ-turnncAA v2synthesis cost
STAGE 05

Family generation, not single sequences

Candidate families are generated around constrained scaffolds: parent sequence, conservative variants, non-natural amino-acid substitutions, cyclized versions, terminal modifications, stability-enhanced variants, and receptor-selectivity variants.

The system searches local neighborhoods around biologically plausible scaffolds instead of sampling the full peptide universe.
STAGE 06

Genetic programming and cross-pollination

Successful families evolve under selection pressure from in-silico scores, with bounded mutation, crossover between motif families, scaffold recombination, and ncAA substitution strategies. Candidates failing safety, synthesis, or fit gates are rejected.

Genetic programming is used as a constrained optimization layer, not as unconstrained random sequence generation.
STAGE 07

In-silico lab assessment

Candidates are scored before any wet-lab handoff: receptor fit, pathway relevance, predicted stability, solubility, aggregation risk, synthesis complexity, off-target concerns, novelty, manufacturability, and assay suitability.

STAGE 08

Human review and wet-lab handoff

Only candidates passing evidence and quality gates move to LabSpace: batch creation, destination lab selection, capability matching, pricing/terms visibility, assay request, status tracking, returned wet-lab results, and feedback into candidate ranking.

Product workflow

From therapeutic intent to testable peptide families.

A linear, auditable path from condition to validated wet-lab batch — every stage produces an artifact a researcher can review.

01

Select condition

Define therapeutic intent or phenotype as the entry point.

02

Map pathways & receptors

Pathways, receptor families, ligand classes, contraindications.

03

BioScout source systems

Evidence-backed mimicry plans from evolved peptide systems.

04

Generate constrained families

Scaffold-bounded candidate families, not single sequences.

05

Visualize peptide–receptor fit

Structure-aware 3D review of binding orientation.

06

In-silico lab assessment

Multi-criteria scoring and rejection gates pre-wet-lab.

07

Prepare wet-lab batch

Hand off through LabSpace to capability-matched partners.

08

Receive results & refine

Wet-lab feedback updates ranking and search direction.

3D structural review

Structure-aware review for receptors, peptides, and fit.

DiscoveryLab includes RealityKit-based 3D visualization on macOS and visionOS. Inspect receptor structures, peptide conformations, binding orientation, candidate families, and interaction-relevant residues.

  • Ribbon / cartoon receptor visualization
  • Peptide backbone and side-chain inspection
  • Candidate switching across families
  • Rotation, zoom, pan with reference axes
  • Tap or select molecular elements for annotations
  • Future support for AlphaFold DB and imported structural references where available
FIG-03 · Receptor 7TM with bound peptide candidateRealityKit · macOS / visionOS
candidate · DL-PEP-0421residues 1–14 · cyclized C-term
H1S2E3G4T5F6T7S8Glu-3 · salt bridge to Lys-202cyclization (i, i+4)x↓ zy
ΔG (est.)−9.2 kcal/mol
RMSD vs. ref.1.34 Å
Contacts11 residues
Evidence-aware tutor

A peptide tutor backed by retrieval and citations.

DiscoveryLab includes a local peptide tutor backed by retrieval, citations, and quality gates. It supports research reasoning around design choices, receptor rationale, peptide families, uncertainty, contraindication concerns, and evidence gaps.

The tutor supports research reasoning; it does not replace expert review, prescriber judgment, or wet-lab validation.
tutor · session DL-2841retrieval · 14 sources · local MLX/RAG
researcher
Why prefer a dual GLP-1R / GIPR scaffold over GLP-1R alone for this program?
tutor · evidence-gated

Dual agonism at GLP-1R and GIPR has shown additive effects on glycemic control and weight loss in published preclinical and Phase 2/3 programs [1][2]. For this condition mapping, family DL-FAM-019 retains GLP-1R primary contact residues while introducing a GIPR-permissive C-terminal motif [3]. Uncertainty: long-term receptor desensitization data is incomplete [4].

[1]Frias et al., 2021 · NEJM
[2]Rosenstock et al., 2023
[3]Internal: family DL-FAM-019 spec
[4]Open question · evidence gap
In-silico lab

Before wet lab: automatic computational triage.

A standardized assessment matrix runs across every candidate. Failing scores constrain or reject; passing scores produce a defensible priority list.

AssessmentWhat it checksWhy it reduces search space
Receptor / pathway relevanceAlignment of candidate to selected receptor and pathway evidenceDiscards candidates with no plausible biological target
Sequence motif plausibilityPresence and orientation of binding motifs and pharmacophoresRemoves scaffolds that violate known structural priors
Stability (predicted)Half-life proxies, protease cleavage liability, oxidation riskFilters chemically fragile sequences early
SolubilityPredicted aqueous solubility under assay-relevant conditionsAvoids candidates that cannot be tested at meaningful concentrations
Aggregation riskβ-sheet propensity, hydrophobic patches, self-association cuesReduces wet-lab failure from precipitation
Synthesis feasibilitySPPS difficulty, cyclization route, ncAA availabilityPrevents prioritization of impractical sequences
Off-target concernsPredicted cross-reactivity to related receptors and proteinsSurfaces candidates needing selectivity engineering
ncAA compatibilityCompatibility of non-natural residues with target chemistryKeeps v2 ncAA expansion within feasible bounds
Assay readinessMatch between candidate and available wet-lab assaysEnsures handoffs are testable as designed
Wet-lab priorityComposite ranking across all assessments aboveFinal gate before LabSpace batch creation
LabSpace handoff

Connected wet-lab collaboration.

DiscoveryLab prepares candidate batches and uploads them to LabSpace — a shared web workspace for selecting wet-lab partners, reviewing capabilities, sending batches, monitoring status, receiving results, and feeding them back into DiscoveryLab.

Researcher accountSign in with Apple
Lab accountSign in with Google
Lab capability profilesassays, instruments, throughput
Pricing & termstransparent per-batch quoting
Batch statusreceived → in progress → results
Returned resultsstructured, attachable to candidates
App notificationsautomatic when results arrive
FIG-04 · LabSpace · batch DL-BATCH-0182partner: Helix Bio Lab #L-228
Batch
DL-BATCH-0182
Candidates
12 sequences · 3 families
Assay
GLP-1R cAMP · cell-based
Status timeline
  1. Created in DiscoveryLab
    Nov 04 · 09:12
  2. Sent to Helix Bio Lab
    Nov 04 · 14:30
  3. Received & QC passed
    Nov 06 · 08:50
  4. Assay in progress
    Nov 07 · 11:00
  5. Results returned
    Nov 12 · expected
Technical credibility

Designed for constrained discovery, not infinite generation.

01Condition-first target selection
02Receptor-conditioned generation
03BioScout mimicry from evolved peptide systems
04Structured candidate families, not single sequences
05v2 non-natural amino-acid support
06Genetic programming with bounded mutation/crossover
07In-silico scoring gates at every stage
08RealityKit molecular review (macOS / visionOS)
09LabSpace wet-lab feedback loop
10Local MLX/RAG peptide tutor for evidence discussion
Search-space math · principle

DiscoveryLab reduces a combinatorial sequence universe into ranked, testable neighborhoods.

Naïve peptide generation scales as alphabet_sizeᴸᴱᴺᴳᵀᴴ. DiscoveryLab reduces effective search by lowering the target receptor set, selecting plausible scaffold classes, applying motif constraints, using structural priors, generating candidate families around plausible neighborhoods, pruning with in-silico assays, and updating future search from wet-lab feedback.

“Every stage either constrains, ranks, or rejects. Nothing is generated simply because it is syntactically possible.”
Reduction operators
  • R₁lower target receptor set
  • R₂select plausible scaffold classes
  • R₃apply motif & pharmacophore constraints
  • R₄structural priors (binding geometry)
  • R₅family neighborhoods around scaffolds
  • R₆in-silico assay pruning
  • R₇wet-lab feedback re-ranking
Research use

Research software with explicit uncertainty.

DiscoveryLab is for peptide research, discovery planning, and candidate prioritization. It is not a substitute for regulatory review, clinical judgment, laboratory validation, toxicology, or controlled studies.

  • no clinical claims
  • no diagnostic use
  • no automated prescribing
  • evidence gaps surfaced explicitly
Begin a discovery program

Move from peptide ideas to constrained, testable discovery programs.

Peptiter · DiscoveryLab · build 1.4.2 · research use only