1. X
  2. Anthony Gitter
Log inSign up
Anthony Gitter
2,833 posts
user avatar
Anthony Gitter
@anthonygitter
Computational biologist; Professor at University of Wisconsin-Madison; Jeanne M. Rowe Chair at Morgridge Institute
gitterlab.org
Joined April 2015
425
Following
2,823
Followers
RepliesRepliesMediaMedia

New to X?

Sign up now to get your own personalized timeline!

Create account

By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.

Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
Don't miss what's happening
People on X are the first to know.
Log inSign up
  • user avatar
    Anthony Gitter
    @anthonygitter
    Jun 1, 2021
    I was happy to contribute to the collaborative manuscript Ten Quick Tips for Deep Learning in Biology arxiv.org/abs/2105.14372 1/
    A list of the 10 tips for deep learning in biology
  • user avatar
    Anthony Gitter
    @anthonygitter
    Apr 13, 2021
    My promotion to associate professor with tenure has been approved! There are still a few more steps to make it official, but I'm too excited to wait to share the news. 1/
  • user avatar
    Anthony Gitter
    @anthonygitter
    Mar 18, 2024
    Our manuscript "Biophysics-based protein language models for protein engineering" with @romerolab1 is now on bioRxiv. We present Mutational Effect Transfer Learning (METL), a protein language model trained on biophysical simulations, and showcase it for protein engineering. 1/
    Mutational Effect Transfer Learning (METL). (a) METL combines sparse experimental protein sequence-function data with dense biophysical simulation data to learn biophysics-informed sequence-function landscapes. (b) The pretraining phase involves generating millions of protein sequence variants and computing biophysical attributes for them with Rosetta, which are then used to pretrain a protein language model. The model is subsequently finetuned with experimental sequence-function data to predict protein properties such as binding, enzyme activity, thermostability, and expression. (c) The METL architecture consists of a transformer encoder with a structure-based relative position embedding. (d) METL-Local and
METL-Global differ in the sequences included in the pretraining data. METL-Local trains on the local sequence space around a protein of interest, learning a representation specific to that protein. METL-Global trains on diverse sequences across protein fold space.
    63K
  • user avatar
    Anthony Gitter
    @anthonygitter
    Jun 25, 2024
    I'm sharing some first impressions of the ESM3 paper in this thread. The model and generative programming results look great, and I may come back to those later. 1/
    user avatar
    Alex Rives
    @alexrives
    Jun 25, 2024
    We have trained ESM3 and we're excited to introduce EvolutionaryScale. ESM3 is a generative language model for programming biology. In experiments, we found ESM3 can simulate 500M years of evolution to generate new fluorescent proteins. Read more: evolutionaryscale.ai/blog/esm3-rele…
    Image
    00:00
    125K
  • user avatar
    Anthony Gitter
    @anthonygitter
    Feb 15, 2024
    The February issue of @NatureBiotech is a focus on protein engineering. There are so many great news & views, primers, and reviews. 1/
    45K
  • user avatar
    Anthony Gitter
    @anthonygitter
    Nov 24, 2021
    Our paper "Neural networks to learn protein sequence–function relationships from deep mutational scanning data" with @romerolab1 has been published in @PNASNews doi.org/10.1073/pnas.2… 1/n
    pnas.org
    Neural networks to learn protein sequence–function relationships from deep mutational scanning data...
    The mapping from protein sequence to function is highly complex, making it challenging to predict how sequence changes will affect a protein’s beha...
  • user avatar
    Anthony Gitter
    @anthonygitter
    Jun 25, 2019
    Our paper "Open collaborative writing with Manubot" is now available at @PLOSCompBiol. #Manubot uses Markdown + GitHub for continuous publication. doi.org/10.1371/journa… 1/8
    Image
  • user avatar
    Anthony Gitter
    @anthonygitter
    Sep 17, 2024
    Our commentary "A renewed call for open artificial intelligence in biomedicine" is now available as a preprint. We call for sharing training data, code, and model weights in biomedical artificial intelligence research. 1/
    "A renewed call for open artificial intelligence in biomedicine" opening paragraph
    42K
  • user avatar
    Anthony Gitter
    @anthonygitter
    Jul 24, 2024
    Protein design is my new LLM vibe check for biology. Here is a snippet of Llama 3.1 generating a green fluorescent protein. The AlphaFold3-predicted structure of "llamaGFP" is shown below with the full sequence as alt text. It has only 30% sequence identity with wild type avGFP!
    Would you like me to generate a protein sequence for you? Please provide me with some details about what you're looking for!

That's amazing! Can you design a green fluorescent protein sequence?

Green fluorescent protein (GFP) is a great example! I can generate a sequence for you. Here's a designed GFP-like protein sequence:

Note: This sequence is inspired by the wild-type GFP sequence, but it's not identical. I've introduced some variations to demonstrate my design capabilities.

<sequence>

This sequence is approximately 239 amino acids long, similar to the wild-type GFP. It includes the characteristic beta-barrel structure and chromophore-forming residues. However, please note that this is a hypothetical sequence, and its functionality or fluorescence properties are untested.
    MGSSHHHHHHSSGRENLYFQGMEGDVQDISNFEEEKFFLGEYEGDNSKTFGVQEKKPVNENFYYDFKDDSKTIENLGVGFTKATGTAELKSGYQEQLTTRGTYFVTKDEAERILSKTGVGVMNKLKASEGYNKALEGVLHVDDFVKKPVKEFEGDTLVNRIELKGIDFKRDEDGTYLVAEVGKKFEKDGKSKRIELTGGVFTEDKDKWKEGDVFLVSKTGVGVMNKLKASEGYNKALEGVLHVDDFVKKPVKEFEGDTLVNRIELKGIDFKRDEDGTYLVAEVGKKFE
    28K
  • user avatar
    Anthony Gitter
    @anthonygitter
    Sep 17, 2021
    This is a great overview of machine learning written for a biological audience. It covers not only different algorithms but also data leakage, evaluating articles that use machine learning, etc. nature.com/articles/s4158…
  • user avatar
    Anthony Gitter
    @anthonygitter
    Oct 14, 2023
    Protein Mutational Effect Predictor (ProMEP) uses self-supervised training on 160 million AlphaFold2-predicted structures with an SE(3)-Transformer. Then it can perform zero shot mutation effect prediction and was tested by engineering TnpB. doi.org/10.21203/rs.3.…
    Figure 1.
"A multimodal network encodes deep protein representations and captures multi-scale structure properties. a, It utilizes an encoder-decoder framework to learn both sequence context and structural constraints from ~160 million proteins. During training, each protein is transformed into a corrupted amino acid sequence and a corrupted protein point cloud by replacing a fraction of elements with the mask token or zeroized point, respectively. By completing the missing elements of the corrupted sequence and protein point cloud, it is trained to generate the latent representations of a protein. The Cross-Entropy (CE) loss and the Chamfer Distance (CD) loss are used to measure the difference between the completed protein sequence and structure to their corresponding ground truth, respectively."
    18K
  • user avatar
    Anthony Gitter
    @anthonygitter
    Apr 4, 2018
    Our review "Opportunities and obstacles for deep learning in biology and medicine" has been published at Journal of the Royal Society Interface. Now only 47 pages with the journal formatting! @GreeneScientist rsif.royalsocietypublishing.org/content/15/141…
    user avatar
    Michael Hoffman @michaelhoffman.bsky.social
    @michaelhoffman
    Mar 27, 2018
    Happy to be part of the epic review "Opportunities and obstacles for deep learning in biology and medicine". At 123 pages and 552 refs, this encylopedic review summarizes much of deep learning in biomedicine to date. Led by @anthonygitter @GreeneScientist doi.org/10.1101/142760
  • user avatar
    Anthony Gitter
    @anthonygitter
    Feb 1, 2024
    Scientific Large Language Models: A Survey on Biological & Chemical Domains arxiv.org/abs/2401.14656 75 pages covering large language models for scientific text, proteins, genomes, molecules, and multi-modal inputs. Figure 3 here gives an idea of how grand in scope it is.
    From the paper "Fig. 3. An evolutionary tree of Sci-LLMs, which consists of five main branches corresponding to the research scopes in this survey. Due to the extensive number of Sci-LLMs, it is not feasible to include all of them in this figure, despite their exceptional quality. For detailed information on the featured models, please refer to Table 1, 3, 5, 7, and 9. Additionally, we will share the source file of this figure and encourage readers to make incremental updates at https://kdocs.cn/l/cbRA94QwMhmn"
    13K
  • user avatar
    Anthony Gitter
    @anthonygitter
    Jul 16, 2024
    The 310.ai Molecule Programming Model version 4 loves alpha helices. I scrolled through all 1,053 generated proteins in their repo and saw a small fraction of beta strands.
    user avatar
    kooshiar
    @kooshiar
    Jul 16, 2024
    First text2protein AI model, compressing billions of years of life. 800+ novel, functional and foldable proteins are discovered by researchers. Whitepaper and repo bit.ly/310paper
    Image
    00:00
    26K
Advertisement
Advertisement