Jun Cheng
@s6juncheng
Followers
1K
Following
206
Media
12
Statuses
143
Computational Biology | Machine learning research scientist at @GoogleDeepMind
Joined December 2016
Excited to share #AlphaGenome, a start of our AlphaGenome named journey to decipher the regulatory genome! The model matches or exceeds top-performing external models on 24 out of 26 variant evaluations, across a wide range of biological modalities.1/6
14
211
911
This is a great idea, congratulations!
Modern GWAS can identify 1000s of significant hits but it can be hard to turn this into biological insight. I'm excited to share our new work combining genetic associations and Perturb-seq to build interpretable causal graphs, out today in @Nature:
0
0
4
AlphaGenome model architecture. follows a U-Net-like structure with an Encoder, a central Transformer Tower, and a Decoder. The architecture has inspiration from Borzoi and AlphaFold.
0
0
3
Excited to launch our AlphaGenome API https://t.co/gL6Je8t2xf along with the preprint https://t.co/ZeLzhR7VBl describing and evaluating our latest DNA sequence model powering the API. Looking forward to seeing how scientists use it! @GoogleDeepMind
5
15
78
Happy to introduce AlphaGenome, @GoogleDeepMind's new AI model for genomics. AlphaGenome offers a comprehensive view of the human non-coding genome by predicting the impact of DNA variations. It will deepen our understanding of disease biology and open new avenues of research.
20
212
1K
AlphaGenome API https://t.co/47wzXZJYIi. Share ideas and feedback in the community forum https://t.co/Kin3ez4xX1, paper https://t.co/UtViwOoXIe. The model source code and weights will also be provided upon final publication.
alphagenomecommunity.com
2
14
32
Having worked on RNA splicing models for years, this is a significant milestone for me. It is a real privilege to work with an extremely talented team. With the new API, I’m excited to see what researchers can find with #AlphaGenome.6/6
1
1
13
The model predicts splicing for 367 human tissues or cell types on both strands. Tissue-specific splicing was predicted with some success, but in general this is still a hard problem. Here we show some examples with mixed results.5/6
1
2
11
One of the most exciting parts of our #AlphaGenome work is the ability to directly predict splice junctions from sequence and also use it for variant effect prediction. This is enabled by modeling the competition between splice sites and junction supporting reads.4/6
2
3
12
This more complete view can help us understand the complex consequences of genetic variants on RNA splicing. It is SOTA on 6 out of 7 splicing variant benchmarks, including predicting splicing variant pathogenicity from ClinVar, sQTL causality, and GTEx splicing outliers.3/6
1
1
6
Mistakes in RNA splicing are a frequent cause of diseases. For the first time, we've built a single model that unifies the prediction of RNA-seq coverage, splice sites, their usage, AND the specific junctions they form, providing a more complete picture of splicing outcomes.2/6
1
3
20
We are looking a PhD Student Researcher at Google DeepMind for 2025 summer! Strong publication record and technical skills on machine learning and genomics preferred. Need to be 80% for a few months. Team in London. Looking forward to hear from you!
1
8
39
It’s been an amazing last couple of weeks, hope you enjoyed our end of year extravaganza as much as we did! Just some of the things we shipped: state-of-the-art image, video, and interactive world models (Imagen 3, Veo 2 & Genie 2); Gemini 2.0 Flash (a highly performant and
213
429
5K
Nice work applying VEP to gene burden testing, bridging variant effect to gene functions.
Really pleased to see DeepRVAT published. A new method for rare variant genetics by @Holtkamp_Eva & @brianfclarke in collab with @gagneurlab. I like to think of DeepRVAT as one network on top of another, harnessing variant annotations from bags of models- AlphaMissense and alike.
0
0
2
AlphaMissense predictions are now more easily accessible thanks to the great work of @emblebi! On AFDB you can also download per protein predictions.
To help scientists explore whether a genetic variant is likely pathogenic, and which regions of proteins are functionally important, we've integrated AlphaMissense data by @GoogleDeepMind into @ensembl, @uniprot & the AlphaFold database. https://t.co/zj87nASrJ5
1
4
16
We hope this update will broaden the potential positive impact of this work and therefore benefit more people. Please reach out at alphamissense@google.com for questions on AlphaMissense. 3/3
0
1
4
AlphaMissense predictions are freely available through https://t.co/JBwvzJMPkW or https://t.co/KmCpw3APLK as before. 2/3
zenodo.org
This repository provide AlphaMissense predictions. For questions about AlphaMissense or the prediction Database please email [email protected]. File descriptions AlphaMissense_hg19.tsv.gz,...
1
2
11
We have received lots of interest in AlphaMissense, the @GoogleDeepmind model predicting the probability of a missense variant being pathogenic, since its release in September. Today we are updating the license terms for AlphaMissense predictions to also allow for commercial use.
8
36
296
Gemini 1.5 Pro - A highly capable multimodal model with a 10M token context length Today we are releasing the first demonstrations of the capabilities of the Gemini 1.5 series, with the Gemini 1.5 Pro model. One of the key differentiators of this model is its incredibly long
185
1K
6K
It's always great to look back on the year in a year-in-review blogpost with @JeffDean & James Manyika. It's been an amazingly productive year for us, doing awesome research, shipping products and advancing science - 2024 is going to be incredible! https://t.co/1lRWKewDue
deepmind.google
This has been a year of incredible progress in the field of Artificial Intelligence (AI) research and its practical applications.
35
136
911