LatchBio

AI protein engineering is now so accessible, my mom could do it. Literally. No code required — just a few clicks. Here are the top 18 models you can use in under 5 minutes:

The data infrastructure speaker lineup was incredible. Thank you all for joining and giving enlightening talks. A recap of highlights from left to right… "Automation isn’t the bottleneck—it’s the capture and structure of quality metadata that slows progress." – Dillon Flood from Elsie Biotechnologies on how scientist-centric data infrastructure helped design a better oligo discovery engine. "We built this to balance the trade-off of usability for scientists and flexibility for developers, while maintaining traceability for entire R&D campaigns"  – Hannah Han L. on building plotting software to replace Graphpad Prism with interactive widgets, reactive cells, and GPU powered visualizations. "Data infrastructure leads not just to faster insights but fundamentally new types of biological understanding." – Kenny Workman on tracing the realistic development of a PCSK9 binding antibody from idea to IND. "We built a comprehensive toolset for biologists on Latch... Turns out the turtle neck guys weren't so bad." – Colin Ng (Kenneth Wang & James McGann) from AtlasXomics Inc. on building next generation bioinformatics tools for spatial multi-omics. "Natural products represent the most validated yet untapped idea in drug discovery" – TJ Bollerman on how Enveda Biosciences is using mass spectra and transformer models to explore nature's chemical space. "Bake in security AND compliance from day one and adopt virtual CISO type tools from the start" – Eddie Abrams from BigHat Biosciences on his top life sciences data infrastructure decisions, good and bad. (Hint: use AWS and python). "Adopt an internal tech radar to guide technology selection and governance. Avoid the "shiny new thing" trap: It can lead to hidden costs." – Jordan Christensen from Recursion, on building an obviously boring tech stack of reliable tools to innovate without unnecessary complexity. "Build a culture that expects data standardization and availability." – David Levy-Booth from Dyno Therapeutics, on building Dyno’s ML platform to solve gene delivery, starting with field-leading CNS capsids. (Note: Look for Liminal ORM on github to keep your Benchling schemas in sync.) "Better tools exist. The bottleneck is adoption." – Jesse Johnson from Merelogic on building a reference map to help biotechs standardize and buy software for specific processes (e.g. experiment design) instead of broad domains (e.g. LIMS). – Fiinally, Robert Policastro, PhD from Ensoma, on balancing robust infrastructure with speed to drive programs toward clinical development. Out of frame :) — Thank you also to the great operators Jordan Ramsay & Emma Krivoshein and stand-up comic MC Blanchard Kenfack. You all crushed it. Keep on building everyone. We'll see you again next year. 👋

Learnings over 3 years building software for drug discovery Over the past three years, I’ve been working with the LatchBio team to build a modular and programmable data infrastructure for modern biotech teams. We’ve built a distributed file system integrated with cloud object stores like S3, enabled structured capture of experimental metadata and integration with Benchling through Latch Registry, supported workflow languages like Nextflow, Snakemake, and Flyte, and built a resizable downstream computer with Latch Pods. But the real difficult set of problems comes with the interactive analysis of data. How do you contextualize assay data and make biological decisions to drive forward R&D milestones? There is a need for a new plotting and visualization tool for the biotech industry. Every drug discovery campaign consists of tens of different assays, each with its own analysis challenges. Take single-cell: Scientists have the best tacit knowledge to perform clustering and cell type annotations for their disease of interest but doing so themselves remains out of reach. Why? Providers like Scale Biosciences and 10x Genomics now handle millions of cells per run, and for datasets of this size, simple operations like clustering alone take hours for every button click. GPU-enabled infrastructure for real-time single-cell analysis is essential but challenging to implement alongside other bioinformatics priorities. Another example is qPCR, where analysis is lengthy: creating Excel formulas to handle plate maps, merging metadata with qPCR readouts, calculating ΔΔCq, and formatting for GraphPad outlier analysis. Data duplication and leakage from copy-pasting across various software further complicate the ability to recover precise analysis steps as companies approach IND. In interdisciplinary teams, bioinformaticians are often tasked with creating visualizations. Still, these requests tend to fall lower on their priority list compared to responsibilities like developing novel bioinformatics algorithms. Bioinformaticians lack the tools to become 10X engineers in their internal organization. There is room for new software that is: ✅ Extremely interactive and customizable for scientists ✅ Programmatic control for developers for more in-depth analysis  ✅ Reusable and integrates seamlessly with upstream components like Registry, cloud storage, and bioinformatics workflows ✅ Versioned and reproducible Incredibly excited to distill these learnings into a new product we’re releasing, Latch Plots. We built Latch Plots to balance the trade-off of usability for scientists and flexibility for developers while maintaining traceability for the entire R&D campaigns.

Had a great time with Kyle Giffin & LatchBio at their 2024 Data Infrastructure for Biotech conference in Boston today. They're working on innovative solutions to help patients get treatments they need - Faster time to discovery, faster time to market. #latchbio #innovation #conference Kenny Workman Alfredo Andere 🦖

Excited to share that AtlasXomics Inc. has been continuing to work with the team at LatchBio to simplify the complexity of spatial biology data analysis, particularly spatial epigenome and multi-omics data. We know how challenging it can be for researchers without programming experience to navigate these complex datasets, which is why we’ve developed an interactive, no-code bioinformatics suite to make it easier for biologists to explore their spatial epigenome data. James McGann and I will be presenting this toolset on October 21st at Latch’s 'Data Infrastructure for Biotech,' where we’ll discuss our vision of making spatial epigenome analysis a part of every research question. Stay tuned for more!

I've been eating, breathing, and dreaming about protein engineering for the past few weeks while building out a new protein engineering toolkit at LatchBio. Check out the blog to see how I used these tools to create a synthetic set of plastic-degrading enzyme "ideas" with structures, sequences, and physical/chemical properties — and how Brontë Kolar used them to design a set of custom proteins to bind to a cholesterol target. https://lnkd.in/dbWBfzKD

Many new machine learning tools for protein engineering have popped up in the last year. In these early days, it might be unclear how to actually use them for real molecular design tasks. After working with academic labs and biotech teams on real problems in research and industry, we’ve assembled a curated toolbox of graphically accessible machine learning tools for protein engineering. We demonstrate how to string them together to build a cholesterol drug and plastic degrading enzymes. To build enzymes we: 1. Finetune ProtGPT2 on enzyme sequences 2. Use finetuned ProtGPT2 to generate a library of novel enzymes 3. Create 3D protein structures with OmegaFold 4. Predict thermal stability with TemStaPro 5. Calculate electrostatic potential and hydrophobicity with PEP-Patch 6. Evaluate aggregation propensity with Aggrescan3D To build PCSK9 binders we: 1. Use Pymol to identify the region of PCSK9 we want to bind 2. Diffuse 10 scaffolds near binding hotspots on PCSK9 with RFDiffusion 3. Use ProteinMPNN to generate 100 sequences for these scaffolds 4. Predict 3D protein structure with ColabFold  This work was spearheaded by Tahir D'Mello and Brontë Kolar. Their intelligence, hard work and passion for science is contagious. Read the detailed flow and reference citations here - https://lnkd.in/grpw-THA.

Excited to share that AtlasXomics Inc. has been continuing to work with the team at LatchBio to simplify the complexity of spatial biology data analysis, particularly spatial epigenome and multi-omics data. We know how challenging it can be for researchers without programming experience to navigate these complex datasets, which is why we’ve developed an interactive, no-code bioinformatics suite to make it easier for biologists to explore their spatial epigenome data. James McGann and I will be presenting this toolset on October 21st at Latch’s 'Data Infrastructure for Biotech,' where we’ll discuss our vision of making spatial epigenome analysis a part of every research question. Stay tuned for more!

Many new machine learning tools for protein engineering have popped up in the last year. In these early days, it might be unclear how to actually use them for real molecular design tasks. After working with academic labs and biotech teams on real problems in research and industry, we’ve assembled a curated toolbox of graphically accessible machine learning tools for protein engineering. We demonstrate how to string them together to build a cholesterol drug and plastic degrading enzymes. To build enzymes we: 1. Finetune ProtGPT2 on enzyme sequences 2. Use finetuned ProtGPT2 to generate a library of novel enzymes 3. Create 3D protein structures with OmegaFold 4. Predict thermal stability with TemStaPro 5. Calculate electrostatic potential and hydrophobicity with PEP-Patch 6. Evaluate aggregation propensity with Aggrescan3D To build PCSK9 binders we: 1. Use Pymol to identify the region of PCSK9 we want to bind 2. Diffuse 10 scaffolds near binding hotspots on PCSK9 with RFDiffusion 3. Use ProteinMPNN to generate 100 sequences for these scaffolds 4. Predict 3D protein structure with ColabFold  This work was spearheaded by Tahir D'Mello and Brontë Kolar. Their intelligence, hard work and passion for science is contagious. Read the detailed flow and reference citations here - https://lnkd.in/grpw-THA.

We’re less than one week out from our one-day conference in Boston! At the Data Infrastructure for Biotech conference, we’ll hear from leaders in the space about their experience designing data infrastructure for biotech: Dyno Therapeutics, Building the lab-in-the-loop data platform for ML-guided design Enveda Biosciences, Unlocking nature's chemistry to find new drugs LatchBio, Leveraging natural language prompts on Latch Plots for real biological use cases Elsie Biotechnologies, Building a modern data infrastructure with the bench scientist at the center Ensoma, Balancing infrastructure with analysis in the drive for clinical data AtlasXomics Inc., Pioneering the next generation of bioinformatics tools for spatial epigenomics We'll see you in Boston soon!