Predict more stable proteins from real experimental data
StrataBio combines high-throughput protein stability experiments with AI/ML to assist protein-engineering teams prioritize the variants most worth testing.
High-throughput experiments → structured stability data → predictive models → data-informed variants
0.0%
Accuracy
Built on small domain (40–80 amino acid) sequence assessed for ddG at room temperature.
- MAE (kcal/mol)
- 0.687
- RMSE (kcal/mol)
- 1.029
- Pearson
- 0.858
- Spearman
- 0.861
Protein Engineering Generates Valuable Data. Too Much of It Is Used Only Once.
Protein-engineering teams generate valuable experimental data across every design-build-test-learn campaign. But those results are often distributed across:
That makes it difficult to systematically learn which sequence, structural, and experimental factors are associated with improved protein stability.
StrataBio is being built to change that.
The StrataBio Biological Intelligence Platform (SBIP) structures high-throughput protein-engineering data and uses it to develop predictive models of protein stability.
Use what has already been learned experimentally to accelerate and inform the next round of protein engineering.
AI can design proteins. Experimental performance still decides what works.
Protein-design models can generate and prioritize enormous numbers of plausible sequences and structures.
Computational confidence is not the same as experimental performance.
StrataBio is designed to feed large, standardized protein-engineering datasets into AI/ML models to close the gap between computational prediction and experimental performance.
But a protein that looks promising computationally may still:
The StrataBio Learning Loop
Every experiment makes protein design and engineering more predictable.
Step 1
Capture
Collect the experiment
Bring together:
- Protein and gene sequence
- Mutations
- Structural information
- Expression system (including host, plasmid, promoter, etc.)
- Assay conditions
- Instrument output
- Measured protein-stability outcomes
- Controls and metadata
Initial Focus: Protein Stability and Durability
Protein-engineering teams already have sophisticated tools for discovering active proteins and optimizing catalytic function. A different question remains:
Which variants are stable enough to justify further development?
StrataBio is initially focused on learning from high-throughput measurements of:
Protein expression
Folding
Thermal stability
Retained folding after stress
Relative stability
Related durability phenotypes
As validated experimental datasets grow, the platform can expand into additional process-relevant conditions, such as solvent tolerance, pH extremes, and salt tolerance.
Better AI Still Needs Better Experimental Data
AI/ML models are rapidly improving at solving complex protein-engineering problems.
But advanced models still depend on:
high-quality, condition-specific, standardized experimental data.
StrataBio is building that experimental-data layer.
Importantly, the platform is designed to capture both:
A variant that does not express, unfolds, loses stability, or fails downstream validation can still provide valuable information about the underlying protein fitness landscape.
Capturing both positive and negative outcomes allows protein-engineering campaigns to become a cumulative data asset rather than a collection of disconnected projects.
From High-Throughput Experiments to Better Variant Decisions
Protein-engineering teams may face hundreds, thousands, or millions of possible variants.
The challenge is not simply generating more candidates.
It is deciding:
Which variants should we test next?
StrataBio will use experimental protein-stability data to narrow the search space and improve that decision.