Protein models need experimental ground truth
AI models can increasingly propose protein sequences, structures, and mutations. But a prediction that looks favorable computationally does not guarantee that a protein will:
- express efficiently
- fold correctly
- remain stable
- maintain structural integrity
- perform consistently when experimentally tested
The StrataBio Biological Intelligence Platform (SBIP) is being designed to connect:
sequence + structure + experimental condition → measured protein stability
and use those relationships to improve future predictions.
The StrataBio Biological Intelligence Platform (SBIP) Data Architecture
The objective is not simply to collect more biological data. It is to create well-labeled experimental data that can support meaningful protein-stability prediction.
Protein Identity
- ·Sequence
- ·Mutations
- ·Protein family
- ·Structural information
Experimental Context
- ·Expression host
- ·Construct
- ·Growth conditions
- ·Relevant transcriptional / translational context
- ·Assay configuration
Stability Measurements
- ·Expression measurements
- ·Thermal stability
- ·Retained folding
- ·Relative stability
- ·Stress response
- ·Half-life
Experimental Metadata
- ·Assay conditions
- ·Controls
- ·Replicates
- ·Batch information
- ·Instrument
- ·Origin
Protein-engineering AI/ML models + task-specific experimental data → more useful predictions
Recent protein-engineering research increasingly demonstrates that general AI/ML models become more useful when they are informed by experimentally measured fitness. StrataBio is being built around that principle.
StrataBio Biological Intelligence Platform (SBIP) is not intended to depend on one permanent protein model. The platform is designed so improved computational approaches can be incorporated over time while the underlying standardized experimental-data asset continues to grow.
Most biological databases are biased toward significantly improved variants.
But unimproved variants, minor increases or decreases in phenotype readout, and failed variants also define the boundaries of a protein fitness landscape.
StrataBio Biological Intelligence Platform (SBIP) is designed to preserve:
- Successful variants
- Unsuccessful variants
- Borderline outcomes
- Uncertainty
- Assay conditions
- Controls
- Batch information
- Origin
The goal is not simply more data. It is:
Better-Labeled Protein Engineering Data
Beyond Stability: Toward Manufacturability
Stability is only one aspect of a protein's commercial viability.
As the platform matures, StrataBio is interested in understanding which measurable protein properties can help inform broader questions of manufacturability, including:
- Expression
- Folding
- Solubility
- Stability
- Durability
- Performance under production-relevant conditions
By building the experimental data needed to understand which measurable protein phenotypes and genotypes are most predictive of successful development and production.
Stability is the starting point
Manufacturability is a longer-term learning objective.
Stability is the starting point
Manufacturability is a longer-term learning objective.