
Excited to share my first contribution here at Illumina! We developed PromoterAI, a deep neural network that accurately identifies non-coding promoter variants that disrupt gene expression.🧵 (1/)

PhD, Illumina AI lab; interested in Deep Learning and genome regulation; also drawing, martial arts, guitar, and death metal! (he/him)

Excited to share my first contribution here at Illumina! We developed PromoterAI, a deep neural network that accurately identifies non-coding promoter variants that disrupt gene expression.🧵 (1/)

1/ I’m thrilled to share a new preprint, where we investigate how long-range sequence-to-function (S2F) models like Enformer, Borzoi and AlphaGenome learn variant effects during post-training: biorxiv.org/content/10.64898…

A quick code drop: I ported Illumina’s PromoterAI from Tensorflow/Keras to PyTorch (github.com/adamyhe/promotera…). I actually wrote about this on the genomicsxai blog a couple months ago (genomicsxai.github.io/blogs/…). This week, I added two major optimizations.

nature.com/articles/s41588-0… Timely review on leveraging deep learning methods for predicting effects of non-coding variants.


Cell-state conditioned sequence model of chromatin & expression trained jointly across diverse cell contexts by @zhou_jingtian & colleagues. Nice addition to the growing collection of joint cis-trans models of gene regulation. Still lots to do make the trans component causal.
Our cells carry the same DNA, but how this DNA is interpreted differs wildly from cell to cell. In a new preprint, Science Fellow @zhou_jingtian and team introduce Bolero, a model conditioned on cell state that predicts chromatin accessibility and expression from DNA alone.

This work is not getting enough attention. These are pretty massive leaps in modeling splicing. Very excited about this. Need to do a deep read but the performance on diverse splicing related tasks looks really good.
#SpliceAI2 splice variant effect benchmarks. Considerably improved performance in 9 evaluations spanning cryptic splicing, sQTLs, MPRAs, population genetics, plasma proteins, and rare disease. 🧬✂️🧵

Take a deeper dive into SpliceAI2 performance!
#SpliceAI2 splice variant effect benchmarks. Considerably improved performance in 9 evaluations spanning cryptic splicing, sQTLs, MPRAs, population genetics, plasma proteins, and rare disease. 🧬✂️🧵

Have you ever looked at a gene in IGV or a genome browser and wondered “What are all those (ATAC-seq/H3K27ac/CTCF ChIP-seq/…) peaks doing?” 🤔 I’m thrilled to share that we developed a new technology called Swap-seq to answer that exact question. biorxiv.org/content/10.64898…

Just put the finishing touches on this beast. 17yrs to get here and the largest, most information dense dataset I had the privilege to generate and analyze. Look for it on bioRxiv next week!

This week we're releasing a preprint exploring state space models for long-sequence genomic prediction, alongside Cerberus, a set of open PyTorch models built on this work. biorxiv.org/content/10.64898…

Check out SpliceAI2, a SOTA deep learning model designed to understand and predict RNA splicing!
Incredibly, incredibly excited to introduce SpliceAI2, the next generation of splicing prediction! 🧬🚀🧵

Excited to announce SpliceAI2: illumina.com/science/genomic… Open model weights: huggingface.co/illumina-ai/S… Source code: github.com/illumina/spliceAI…

Illumina's genomic AI models now enable researchers to identify 2x as many variants with predicted biological impact. Identifying these variants and understanding their downstream effects can unlock new biological insights to advance rare disease and drug discovery research. The newest generation of our leading-edge genomic AI models are designed to help researchers assess the functional impact of splice, promoter, and missense variants. With SpliceAI2, out today, we aim to help researchers advance rare disease research. In the words of the vice president of BioInsight AI Lab, Kyle Farh: “Illumina is advancing AI to systematically shrink the portion of the genome that remains uninterpretable.” 🔗 Learn more: assets.illumina.com/content/… For Research Use Only. Not for use in diagnostic procedures.

Super excited to share SpliceAI2! ✂️🧬 It predicts splice-site and splice-junction usage, reconstructs complete transcripts, and models tissue- and disease-specific splicing. Proud to have contributed, and huge credit to the incredible team behind it! Check out the tweetorial!⬇️
Incredibly, incredibly excited to introduce SpliceAI2, the next generation of splicing prediction! 🧬🚀🧵

Incredibly, incredibly excited to introduce SpliceAI2, the next generation of splicing prediction! 🧬🚀🧵