Antibodies
Clinically established, modular recognition molecules with powerful effector and engineering options, but greater sequence and paired-chain complexity.
High translational relevanceASCENT · Open-source molecular recognition
ASCENT stands for Adaptive Systems for Computational Engineering of Novel Therapeutics.
Designing molecules that recognize the right target.
We are starting to build ASCENT as an open-source effort for experimentally grounded AI design of antibodies, nanobodies, and de novo miniproteins. The platform connects generative design, structural prediction, high-throughput screening, NGS analysis, and model retraining in a unified Design–Build–Test–Learn system.
AntibodiesNanobodiesMiniproteins
Antibodypaired-chain recognition
Nanobodycompact single domain
Miniproteincompact de novo recognition
Recognition architectures
One molecular problem. Three design formats.
ASCENT is organized around the shared problem of molecular recognition rather than a single scaffold class. Each architecture contributes different advantages in geometry, stability, manufacturability, and translational use.
Nanobodies
Compact single-domain binders that are readily encoded in pooled libraries and support rapid experimental feedback on affinity, specificity, and stability.
Initial high-throughput testbedMiniproteins
Compact de novo proteins whose folds and binding surfaces can be co-designed for stable, target-specific molecular recognition, with precise control over geometry and presentation.
Programmable compact recognitionPlatform architecture
Format-agnostic at the core. Format-aware where it matters.
ASCENT shares target representations, structural evaluation, experimental readouts, and learning infrastructure across recognition formats while retaining specialized models for each molecular architecture.
Target and epitope definition
Represent the desired surface, geometry, biochemical environment, and specificity constraints.
Architecture-specific generation
Generate antibody loops, nanobody variants, or miniprotein backbones and binding surfaces using architecture-appropriate generative and sequence-design models.
Unified prediction and filtering
Evaluate complex geometry, folding confidence, interface quality, sequence plausibility, stability, and developability.
Shared experimental learning
Integrate binding, specificity, affinity, expression, protease resistance, thermal stability, and NGS enrichment into retraining and benchmarking.
Design–Build–Test–Learn
Experiments become model improvements.
ASCENT is designed to learn from successful binders, weak binders, nonspecific sequences, unstable designs, and complete failures—not only from structures that were eventually solved.
01
Define
Select a target, epitope, molecular format, and performance objective.
02
Generate
Sample backbones, interfaces, loops, and sequences across one or more architectures.
03
Predict
Rank folding, complex geometry, confidence, specificity, and stability.
04
Measure
Screen pooled libraries for binding, affinity, specificity, expression, and stability.
05
Learn
Retrain models, compare workflows, and release reusable datasets and benchmarks.
Initial experimental focus
Start where feedback can scale.
Early ASCENT campaigns will prioritize nanobodies and miniproteins because both can be encoded through pooled oligonucleotide synthesis and evaluated at high throughput. These systems provide rapid, information-rich feedback for improving generative design and prediction before extending the same learning framework to more complex paired-chain antibodies.
Pooled synthesis of thousands of designed sequences
Affinity-ranked yeast-display selection
Specificity and off-target counterselection
Protease and thermal stability screening
NGS-based enrichment and failure analysis
Open outputs
A community resource for molecular recognition engineering.
ASCENT will complement existing prediction and design tools by adding standardized experimental integration, cross-format benchmarking, iterative retraining, and accessible deployment.
Experimentally grounded datasets
Open measurements linking target, sequence, predicted structure, affinity, specificity, expression, stability, and screening enrichment—including unsuccessful designs.
Improved prediction and generation
Model checkpoints and training pipelines for molecular recognition across antibodies, nanobodies, and miniproteins.
Reproducible closed-loop engineering
Open benchmarking, visualization, NGS analysis, and deployment tools for comparing and refining Design–Build–Test–Learn campaigns.
Collaborate with ASCENT
Help build an open foundation for programmable molecular recognition.
ASCENT is intended as a shared ecosystem for model developers, protein engineers, experimentalists, and researchers applying recognition molecules in therapeutics, diagnostics, delivery, infectious disease, and biomolecular materials.
ASCENT
Adaptive Systems for Computational Engineering of Novel Therapeutics
An open-source platform for experimentally grounded AI design of molecular recognition.