New preprint: clear signs that variant effect predictors are now shaping ClinVar labels.
Good for clinical use, but it creates circularity: predictors are trained, benchmarked and calibrated on labels that predictors helped create. https://t.co/GElUz4D7Vq
In our new Nature Communications article, we identify and discuss factors that can lead to discordance between computational variant effect predictors and functional assays.
https://t.co/ZjJNrcsiTK
New preprint! We used deep mutational scanning to measure the effects of amino acid substitutions in ERCC2.
Our assay is strongly mechanism-selective, showing better discrimination between different disease phenotypes than any computational predictors. https://t.co/4e5Sbh2pzN
Out today in PLOS Comp Biol:
https://t.co/azjJ1w8955
Intrinsically disordered regions make variant prediction deceptively easy for benign changes but very hard for pathogenic ones. Our work shows why current tools struggle here, and why disorder-aware approaches are needed.
Genetic variants🧬 can cause protein-coding mutations resulting in disease. What about in alternative isoforms? Excited to share my 1st first-author pre-print! Here we show many variants reside in tissue-specific isoforms, causing coding mutations. A🧵 1/ https://t.co/fWbeBAsuUY
New paper out in Genome Biology! 🎉
We lay out best-practice guidelines for releasing variant effect predictors, developed through the Atlas of Variant Effects Alliance @varianteffects
Open, interpretable, and clinically useful VEPs are the goal.
📄 https://t.co/YUsRgE4usw
New review out with @L_Geras@teichlab ! 🧬🔬 We explore how protein structure is improving variant effect prediction. Advances in structural modelling, including AlphaFold, are powering methods like AlphaMissense, PrimateAI-3D, and CPT-1, leading to better accuracy and interpretability. Read more: https://t.co/ywUNAYv0gH
We are hiring a new postdoctoral bioinformatician!
Come and work with us on multidisciplinary cancer genomics projects at the Institute of Genetics and Cancer in Edinburgh:
https://t.co/CrT8MuLQee
https://t.co/RHIDO7Npor
https://t.co/5z3Y7KxJG5
Protein predictive scores which predict the likelihood that the protein is associated with a dominant-negative, gain-of-function or loss-of-function mechanism are displayed. Curated literature support for a molecular disease mechanism is also shown. @jmarshlab@BdonyiMihly
Are you interested in AI and structural biology, genetics and the non-coding (aka “dark”) genome?
A PhD studentship through the UKRI AI and biomedical innovation programme at @EdinburghUni@EdinUni_IGC has become available with myself and prof @jmarshlab