Engraving of a radiant sun over a calm horizon.

Inserloft ResearchProject ProteusModel 001

CNB-1

Caelis Neural Base 1, the first experimental model for Project Proteus, explores biological sequence space and structured protein generation.

StatusExperimental / Research
ProjectProject Proteus
OrganizationInserloft Research
Model roleFirst experimental Proteus model

Overview

A first signal from the search space.

CNB-1 tests whether a compact system can produce biologically interpretable outputs rather than arbitrary sequences. Its outputs are computational candidates for inspection and stronger testing, not validated proteins or therapeutics. Generation starts a research loop.

Experimental output

Structured biological generation.

An early experiment produced a candidate of about 350 residues with a hydrolytic-enzyme objective.

  • 01

    Sequence scale

    About 350 amino acids.

  • 02

    Target class

    A designed hydrolytic enzyme linked to PET degradation research.

  • 03

    Seed

    MNPAQ, a signal-peptide motif from natural Ideonella sakaiensis PETases.

  • 04

    Research target

    Computational PET hydrolysis, with terephthalic acid and ethylene glycol as intended products.

Biological hypothesis

From sequence to testable hypothesis.

The candidate was designed around PET-hydrolase behaviour. A Ser–His–Asp catalytic triad and an alpha/beta-hydrolase fold serve as validation checks: design criteria, not evidence of activity.

Next come folding, active-site geometry, stability, substrate binding and catalytic activity.

Benchmark

Protein-model evaluation.

Benchmark scores for CNB-1 and comparison models
BenchmarkCNB-1ProGen 2-5ProtGPT-2ProtGPT-2ESM-1v
ProteinGym.48.43.35.35.45
TAPE (Fluor).68.65.58.58.67
TAPE (SecS).62.62.55.55.64
CATH (Acc).55.55.48.48.57
FoldSeek.58.58.48.48.59

Interpretation

What the benchmark does, and does not, establish.

Scores describe performance on specific tasks under specific protocols. They are not proof of biochemical function, therapeutic utility or experimental success.

Research architecture

Generation is only one layer.

  1. 01Representation
  2. 02Generation
  3. 03Evaluation
  4. 04Iteration

Limitations

CNB-1 is an experiment, not a finished biological system.

It is a prototype. Planned work:

  • Structure prediction
  • Active-site analysis
  • Folding and stability
  • Substrate interaction
  • Negative controls
  • Reproducibility
  • Experimental validation

Next work

Build → test → measure → learn.

Expand the model, improve representations, strengthen evaluation, and explore larger candidate spaces.

PROTEUS / CNB-1
  status: EXPERIMENTAL
  output: STRUCTURED BIOLOGICAL CANDIDATES
  first documented target: PET HYDROLYTIC ENZYME
  sequence scale: ~350 aa
  evaluation: COMPUTATIONAL BENCHMARKS
  next: STRUCTURE → FUNCTION → VALIDATION