Open Research
Preprints documenting the models behind the Orbion platform. Every paper includes methods, benchmarks, and evaluation against public baselines — shared openly for the community to review and build on.
April 22, 2025
Can Quantum Computing Actually Improve Protein-Ligand Binding Prediction?
Quantum computing is unlikely to replace classical protein-ligand prediction pipelines anytime soon, but it may still help in specific cases. In this experiment, quantum-derived descriptors gave modest overall gains, yet showed much stronger value for ligand classes with richer electronic structure, such as porphyrins, nucleotides, and cofactors. The main takeaway is that quantum methods look more promising as a targeted add-on for hard binding problems than as a general solution today.
April 15, 2026
Beyond Structure and Affinity: Context-Dependent Signals for de novo Binder Success
De novo binder design still fails often in experiments, and structure or affinity scores alone do not predict success well. By re-analysing two public benchmarks with biology-informed sequence features, we found signals that transfer across settings as well as others that depend on binder format and assay context. These results suggest binder screening should be more context-aware and multi-factor, helping catch likely failures earlier.
November 11, 2025
AstraBIND: Graph Attention Network for Predicting Ligand Binding Sites
AstraBIND predicts ligand classes and binding residues by combining sequence, structure, and homology in a lightweight graph neural network. It is designed to be faster and more scalable than heavier structure-based methods while still giving strong performance across many ligand types. This makes it practical for rapid in silico screening and integration into protein design workflows.
October 4, 2025
AstraPTM2: A Context-Aware Transformer for Broad-Spectrum PTM Prediction
AstraPTM2 predicts 39 post-translational modification types across full-length protein sequences, helping overcome the usual tradeoff between context, coverage, and rare-PTM performance. By combining sequence embeddings, structural features, and protein-level signals, it captures both local motifs and long-range effects. The result is a well-calibrated model for broad PTM prediction that also supports practical experimental planning through the Orbion platform.
August 8, 2025
AstraROLE2 & AstraSUIT2: Multi-Task Annotation Models for Functional Profiling of Proteins
AstraROLE2 and AstraSUIT2 are multi-task protein annotation models that give a broad functional profile in one pass, instead of forcing researchers to combine many separate tools. Together, they predict features such as function, pathways, cofactors, domains, localization, and membrane properties from protein sequence-derived representations. The models achieved strong benchmark performance, suggesting they can provide fast, high-quality annotations for protein characterization and hypothesis generation.
Have a question?
Whether you want to understand our models, test a specific protein, or just ask how Orbion applies to your research — send us your question and we'll get back to you. Or, if you'd prefer a live walkthrough, book a demo.
