LatchBio

How a hidden tab in an Excel spreadsheet wiped $12 billion off Amgen’s market cap. Nine months ago, Amgen published data from a phase 1 trial of their obesity drug prospect, MariTide, in Nature Metabolism. Unbeknownst to many, the Excel spreadsheets shared alongside the publication contained hidden tabs with crucial bone density data. An analyst at Cantor Fitzgerald discovered these hidden tabs, which included data on femur, lumbar, and hip DEXA bone density scans. One tab showed that patients on the highest dose of MariTide lost about 4% bone mineral density over 12 weeks, suggesting potential risks with GIPR antagonists like MariTide. Following the revelation, Amgen's stock plummeted 7%, wiping out $12 billion in market cap. Jefferies analyst Michael Yee called the bone density concerns a "non-issue," noting the data was inconsistent and based on a small sample size. Amgen stated they saw no association between MariTide and bone density changes and remained confident in their drug... In scientific research, transparency and clarity is crucial. Excel's lack of traceability and the ease of missing data can lead to significant misunderstandings and massive financial repercussions. Amgen anticipates the phase 2 topline data for MariTide later this year, I imagine they won't be using the hidden tabs feature this time around.

Dyno Therapeutics uses machine learning guided engineering of large molecular libraries to build new AAVs with: - 100x greater brain transduction - 10x improved liver detargeting - comparable production Inspired by their work, Hannah Han L. replicated?a historic paper by Ogden?et al,?which took a look at?all single-codon mutants of the AAV2 cap gene across 735 positions. She reproduced some of the key scientific results and had fun thinking about the same biological questions the Dyno team was confronted with years ago: 1/ How do mutations effect viral production? 2/ How do mutations effect biodistribution? 3/ How can we use machine learning to generate new, optimal designs? Along the way she digs into the guts of interesting visualizations and shows how to use Latch Plots — a Python-based scientific plotting framework - to answer these concrete questions.

A very special night in San Francisco bringing together the biotech community for the first stop on the Founder-Led Biotech Tour. A huge thank you to our SF co-hosts - Kevin Parker, Seemay Chou, Josh Moser, David Schaffer, Brandon Wilson, Ph.D., Alfredo Andere ??, Ashley Zehnder, Lexi Rovner, John Suliman, Armand Cognetta, Nathaniel Chu, Lada Nuzhna, Raphael Townshend, Eerik Kaseniit, Brandon White, Nicholas Larus-Stone, Nicolas Tilmans, Egan Peltan, Michael Becich, Adil Yusuf, Richard Yu, Tim Schnabel, Tess Bevers, Jacob Borrajo, Ashton Trotman-Grant, Ivana Muncie-Vasic, Janine Sengstack,Trevor Martin and Jasmin Hume, PhD. Thank you to Future House for hosting and for our wonderful sponsors for making this all possible. Global Sponsors: J.P. Morgan, Wilson Sonsini Goodrich & Rosati, Eli Lilly and Company. City Sponsor: LatchBio Founder-Led Biotech upcoming stops: NYC 11/14 London 11/18 Boston 11/20 Register here: https://founderledbio.com/

Many legitimate therapeutics companies said "no" to Latch early on. These are emails I received in May of 2022 from lost clients. We have since listened intently to their feedback, and built what I think is a genuine solution to a huge problem in biotech. It is called Latch Plots. ?? ?? -- For context, in 2022, our product was based on interviews with 100+ bioinformatics and computational teams in biotech. We knew data & computing infrastructure for these teams was important, but missed a major problem the broader biotech organization faces: scientific data interpretation. When a bench scientist (a bioengineer, chemist, molecular biologist, geneticist, or immunologist) receives data from a computational team (a data engineer, data scientist, or bioinformatician), they have questions. A ton. "What is the meaning of this data? How does it work? Can I tweak the plots? What other samples can I compare here? What are the axes? What do these genes do? How can I best visualize what's going on? Can we re-run this with different parameters?" Typically the plots are hard to change, so developers hear these questions, go back, change the code, tweak the statistics, and work on making new plots. Meanwhile the scientists (who really want to do their own analysis) are just waiting... This creates a huge bottleneck and slowdown in the scientific method of observation, hypothesis, and conclusion. Scientists want to conclude independently, in minutes. Not wait for you to write code. If developers want to iterate on code, and wet lab scientists want to iterate on plots, you inevitably get teams working in siloed and disjointed systems, doing their own thing in a way that is slow and impossible to track. We noticed that the best teams in biotech have fast, iterative feedback loops between wet <> dry lab. These teams focus on offering scientists self-serve insights with statistical integrity. Scientists are then empowered to answer questions on their own. It turns out this is super valuable - just really hard to do with existing tools (Graphpad Prism, Jupyter, RShiny, Plotly, etc.). This is the problem our customers wanted to solve in 2022... Hence the emails. That is why I'm so psyched about the launch of Plots. And incredibly proud of the latch engineering team for what they shipped. Latch Plots is a hybrid graphing software for programmers and non-programmers in biotech to generate scientific conclusions, together. Developers can write code in cloud-hosted notebooks, configuring flexible visualizations for scientists. Anything can be created, from differential expression, pathway enrichment, and single-cell clustering and annotation, to large scale statistical GWAS studies. I'm excited to see how people will push the boundaries with this tool. And hopeful this improves the communication and collaboration between scientists and developers. If it doesn't, let us know - we'll listen to your feedback.

Huge shout out to the eng + product team at Latch, they have been shipping like crazy! Here is a recap of just the past month: 1) Assembled a Protein Engineering Toolkit with 16 user-friendly pipelines. Tools like RFdiffusion let you design new protein structures. We did a case study building plastic-degrading enzymes and cholesterol drugs. 2) Launched Latch Plots—a new way to visualize biological data. Traditional tools like graphpad weren't cutting it, so we built something ourselves. It's no-code + python-based with an AI integration, so you can generate plots just by describing what you need. 3) Used GPUs to speed up single-cell data analysis. Tasks that used to take over 30 minutes now run in under a minute. It's pretty game-changer for handling large datasets within a notebook. 4) Also on the GPU front, an accelerated version of nf-core/methylseq for epigenetic analysis. With data sizes exploding (1M+ cells), using GPUs in bioinformatics workflows is becoming essential, and we're seeing significant performance boosts. 5) The second biotech data infrastructure conference where engineers and scientists shared ideas openly. It was amazing to see so much collaboration. We're already planning to do it again next year. And they are already iterating on the next set of shipments with users, looking forward to continue watching their pace.

October was busy.

Read our latest Product Newsletter to see what we've been working on over the past month! This edition highlights our: -New suite of molecular designs tools for protein engineering -Replacement for GraphPad Prism + Excel, Latch Plots -Roadmap for GPU-enabled bioinformatics -Recent open conference for data infrastructure in Boston

Innovation in antibiotic development has stalled since the Golden Age of soil screening in the mid 1900s. New computational techniques offer a path forward. We walk through a realistic virtual screening campaign to find molecules that target gram negative bacteria. And predict ADMET properties + binding affinity for 250K molecules to identify a realistic lead. Our virtual screening flow consists of the following steps: 1. Downloading 250k compounds from CartBlanche22 2. Predicting ADMET properties with ADMET-AI to arrive at 630 candidate compounds 3. Converting the SMILES representations of the compounds to 3D structures 4. Using GNINA to predict docking affinity values against our selected target We recovered a handful of realistic leads, including a molecule ZINCn500001tBjEc with desirable toxicity and affinity properties. All tools used in this workflow (and more, like DiffDock) are publicly available on LatchBio meaning you can choose your own target and replicate this flow. This project was entirely conceived and executed by Bront? Kolar, a bioinformatics engineer at Latch. Shoot an email to [email protected] to learn more and work with her on your next project.

We are setting out to build new tools to analyze, graph and present scientific work. GraphPad Prism is a staple of the computational workbench, yet it has not kept pace with the increasing demands put on scientists by modern assays. Users struggle to integrate data from their other platforms, dry-lab scientists lack access to the specifics of statistical analyses and plotting (code), and a lack of collaborative editing and shared workspaces hinders teamwork. Latch Plots will focus on serving wet-lab scientists through: — Integrations with existing data. Excel, Benchling, GDrive, S3, GCP, SRA, BaseSpace, drag and drop, CLI, or Benchling. — Intuitive GUI widgets. — No-code plots (scatter, bar, heat-maps, volcano, etc...) and statistical analysis. Always backed by accessible and editable Python code. — Built-in natural language models like GPT4 and Claude. — Dashboards created by bioinformatics workflows. For example, an nf-core/rnaseq workflow generating a DESeq2 dashboard.

A lot is happening at the intersection of AI and Science and we at Felicis are excited about all the developments. 1. The Nobel Prize in Chemistry this year went to David Baker, Demis Hassabis, and John Jumper of Google DeepMind for their work in computation?protein design and prediction. Similarly the award?for Physics went to John Hopfield?and Geoffrey Hinton for their work with ML and neural networks. This is the first time in history that the Nobel prizes have been awarded?to ML focused researchers in both disciplines. 2. Billions of dollars has been (and will continue to be) invested into computational biological design companies like Xaira, EvolutionaryScale, Profluent, Chai Discovery, Inceptive, and others. 3. Jensen Huang has spoken about how digital biology is the?next frontier for machine learning. He mentions it is engineering not science. And that we can finally represent the language of biology and chemistry as an engineering problem, rather than a biological one. After years of backing companies like Guardant Health, Ginkgo Bioworks, Inc., Recursion, BioAge Labs, Orca Bio, with founders from organizations like the Broad Institute of MIT and Harvard, Scripps Research, and Stanford University School of Medicine, we are excited to continuing to further our commitment to applications?in the sciences, by supporting the next generation of researchers. Felicis is proud to be?a?donor to?the?Arc Institute, a?multidisciplinary biomedical science organization founded to study complex diseases like cancer, neurodegenerative?disease, and immune dysfunction.??We are excited to watch Arc’s progress at the interface of biomedicine and machine learning, and look forward to building a relationship with other biotechnology focused research groups. PS. Please join us for our next AI x Bio event highlighting?the work of companies like?Chai Discovery,?Angstrom AI,?LatchBio, with assistance from?Lux Capital. https://lu.ma/nm63eeik Thanks to Silvana Konermann & Michael T. for your partnership!