New in Briefings in Bioinformatics: we use the Gene Ontology @news4go as a logical framework to ask whether genome-scale function annotations are biologically plausible, not only accurate protein by protein. #GeneOntology #Bioinformatics 1/6
We applied the framework to curated annotations from six model organisms and predictions from seven methods. Curated annotations mostly satisfy the constraints; current predictors do not yet produce complete, coherent and consistent genome-scale annotation sets. 5/6
Our ontology-aware metrics are complementary to CAFA-style protein-level accuracy.
Paper: doi.org/10.1093/bib/bbag336
Code/data: github.com/bio-ontology-rese…
#CAFA #FunctionCOSI #ProteinFunction 6/6
New in Bioinformatics: INDIGENA, an inductive method to predict disease-gene associations by ranking genes for a set of phenotypes. 🧵 1/5
A supervised signal from known gene-disease associations makes the embeddings and the similarity measure task-specific. On mouse models of human disease it beats inductive semantic-similarity baselines and matches transductive methods, while staying general. 4/5
Work led by Fernando Zhapa-Camacho.
Paper: doi.org/10.1093/bioinformati…
Code: github.com/bio-ontology-rese…
5/5
#ESWC2026 paper 3/3 — Robust KG Embedding via Denoising: unified robustness framework treating KGE scoring as energy-based, exploiting the denoising / score-matching link. Adds certified robustness metrics via randomized smoothing. Code: github.com/tewiSong/RKGE
#ESWC2026 paper 2/3 — GeometrE: fully geometric multi-hop reasoning on KGs. Every logical operator maps to a pure geometric transform in latent space. Plus a transitive loss preserving r(a,b) ∧ r(b,c) → r(a,c). github.com/bio-ontology-rese…
3 papers from our group accepted at #ESWC2026 in Dubrovnik next week! Excited to present work on drug-safety knowledge graphs, geometric multi-hop reasoning over KGs, and robust KG embeddings via denoising. See you in Dubrovnik!
SIDEKICK: a knowledge graph of drug indications, contraindications & adverse reactions from 50K+ FDA labels via LLM + Graph-RAG, mapped to HPO/MONDO/RxNorm. Beats SIDER & OnSIDES on side-effect-based drug repurposing. At #ESWC2026, Dubrovnik. sidekick.bio2vec.net
🧬 Proud to share our new paper! A complete genome assembly from Saudi Arabia (KSA001), freely available to all. Special shoutout to @coolmaksat for leading the bioinformatics. Also hard work to get this IRB-approved nature.com/articles/s41597-0…
2nd day of #NeSy2024 is about to start, with a session on neurosymbolic languages. Will be great to see how this work can contribute to reproducibility of NeSy research and experiments.
"Size matters" --- knowledge graphs and ontologies in life sciences are possibly larger than in any other field; they are a formidable challenge for any AI method, symbolic or neural #nesy2024
Billions of statements in biological knowledge graphs! Millions of nodes, billions of triples.
Neurosymbolic methods are not yet ready to handle this data. (From @CPesquita's keynote at #NeSy2024)