MiraOmics

Get ready to watch and listen to Eugene Bolotin, Ph.D. discuss the MiraTyper from MiraOmics and how to simplify your cell typing!

🚀 Exciting Results: MiraTyper from Miraomics Matches or Exceeds Leading Cell Typing Embedding Models We’re thrilled to share that we have benchmarked our MiraTyper embeddings against the Chan Zuckerberg Cell Typing Leaderboard — and the results speak for themselves. Using both the CZ benchmark dataset and our internal evaluation framework, MiraTyper matches or exceeds the performance of the top public embeddings on the most relevant part of the benchmark (logistic regression). But performance is only part of the story. The CZ benchmarks test embeddings only — the mathematical “engine” that represents cell properties. However, in practice, you can’t drive an engine alone. To classify cell types, you need both the engine (embedding) and the vehicle (classifier) that interprets it. MiraTyper combines both. Our embedding is state-of-the-art — but our integrated classifier, trained on over 3.5 million cells, takes it further by delivering ready-to-use, high-accuracy cell type predictions right out of the box. No need to build the rest of the car! That means: ⚙️ High performance: Benchmarked,  state-of-the-art embedding performance 🧠 Complete package: A powerful built-in classifier for actual annotation 🏢 Private and easily deployable: Immediate on-premise deployment — no GPU, no complex setup 👩🔬 Ready to go: No need for a dedicated bioinformatician to convert embeddings into labels 💡 Customizable: Tailor models to specific client datasets and use cases If the embedding is the engine, MiraTyper is the car — built, tuned, and ready to drive. —------------------------------------------------------------------------------ As a reminder: ☕️ For less than the price of a coffee, you can now try MiraTyper on your own data on LatchBio and examine it in detail on Pythia BioSciences platform. Reach out to us to find out how. Links to the benchmarking tools by CZ and how to try MiraTyper are in the comments:

🎉 Exciting News: MiraTyper from Miraomics is Now Publicly Available on LatchBio! After months of development, we're thrilled to announce that MiraTyper is now accessible to the research community for automated cell type annotation at unprecedented resolution. What is MiraTyper? Amazing cell type prediction for your single cell RNA-seq data! Just upload your scRNA-seq, H5AD file and get highly detailed cell type annotations in under 10 minutes. Getting Started is Easy: 1️⃣ Create a free LatchBio account and add payment info 2️⃣ Find MiraTyper in public workflows 3️⃣ Upload your h5ad file 4️⃣ Get results in <10 minutes Why MiraTyper? ✅ >80% hierarchical precision and recall on 10x genomics generated human data. 🔬Batch resistant 🔬Tissue agnostic 🔬Works on spatial data (Visium, Xenium, CosMx) 🔬Can work across organisms (through homologous gene mapping). Demo Version Details: 🔬 Files up to 100k cells 🔬 Data randomly subsampled to 50% for demo (e.g., 10,000 → 5,000 cells) Full Version Details: 🔬No demo limitations 🔬Local install available through docker or python package Customization through professional services: 🔬Unusual model organisms. 🔬Rare or unusual cell types. 🔬Custom disease tuning (cancer, other) We can adapt the model for your needs! We Want Your Feedback! We're actively collecting bug reports and user feedback to improve MiraTyper. Don't hesitate to reach out, through linkedin, or send an email at eugene@miraomics.bio, if you have comments, suggestions, custom requirements. #SingleCell #Bioinformatics #ScRNAseq #CellTyping #Genomics #ComputationalBiology #ResearchTools Special thanks for RJ Honicky for driving this project and Pythia Biosciences for providing 10TB of training data! Follow these steps on LatchBio after making the account.

🧬 Exploring the Significance of Benchmarking with Hierarchy in Cell Typing When we built our Cell Typer -MiraTyper, we decided to evaluate it against state of the art methods. However, we quickly run into a problem! Our accuracy was really low (F1, precision, recall) ! However, when we looked closely into our results, we were calling cells at a very fine level and beating the competition. For example: we labeled "Naive CD4 T-cells" vs competition labeled "CD4 T-cell". What happened? ⚠️ The Pitfall of Standard Metrics Conventional evaluation methods tend to treat all classification errors uniformly, overlooking the nuanced biological implications of misclassification scenarios. 🌿 Embracing Hierarchical Benchmarking Hierarchical benchmarking introduces a paradigm shift by leveraging the Cell Ontology to assess the biological proximity of predictions. This approach not only acknowledges near-miss predictions but also rewards their significance in real-world applications, offering a more nuanced evaluation of model performance. 📊 Performance Insights from Miraomics In the analysis of brain and immune datasets, Miraomics showcases its prowess: - Hierarchical Recall: 0.97 (compared to 0.11 by CellTypist) - F1 Score: 0.98 (in contrast to 0.19) - Lightning-fast processing under 1 minute for 30k+ cells - Streamlined operations with no need for manual model selection or label harmonization ⚡ Don't Miss the Biological Context Relying solely on flat accuracy metrics for model evaluation overlooks the critical biological context that hierarchical benchmarking can unveil. Stay attuned to the nuances of cell typing with a comprehensive approach like hierarchical benchmarking.

🚀 Presenting Miraomics Cell Typer - MiraTyper: Revolutionizing Cell Type Annotation 🧬 Miraomics is reshaping the landscape of single-cell analysis with our cutting-edge Miraomics Cell Typer. In a recent benchmarking study against established tools like CellTypist. The outcomes speak volumes: ✅ Achieved a remarkable 10x higher recall in brain datasets ✅ Demonstrated a 2x higher recall in immune datasets ✅ Attained an impressive up to 98.4% hierarchical F1 score ✅ Executes in less than 1 minute for over 30,000 cells 🔍 What makes us stand out? Our platform is powered by a hybrid generative AI model, trained on a vast dataset of over 100 million cells from the CellXGene collection. Key distinguishing features include: - Cluster and batch agnostic design, eliminating the need for manual clustering - Pre-harmonized labels aligned with CL ontology, requiring no post-processing - Adaptable and customizable to any dataset and tissue While others grapple with manual model selection and label normalization, Miraomics ensures swift and consistent results with exceptional accuracy. Whether unraveling immune cell signaling intricacies or mapping the human neocortex, our technology keeps pace with the complexities of biology. 🧪 Curious about our performance on your specific data? We provide tailored benchmarking services for clients to witness the impact firsthand. 📊 Explore the comprehensive performance breakdown, visual representations, and more detailed insights in our latest benchmarking report. 🔗 Connect with us to learn more: info@miraomics.bio or reach out to us on LinkedIn! Additional details: https://lnkd.in/gW9ncGYr #singlecell #celltyping #genomics #biotechinnovation Miraomics RJ Honicky

We are incredibly excited to see where we will take AI in the future!

We’re thrilled to announce that 30 million single-cells have just been released through our partnership with LatchBio and Pythia Biosciences! This marks a major step forward in miraomics’ mission to help scientists truly understand their data. By making this massive dataset freely and easily accessible, we are: 🔓 Breaking down the silos that fragment biological research 🌍 Placing our clients' data in the broader context of global biological knowledge 🧠 Empowering discovery by connecting isolated experiments to high-quality, richly annotated public datasets At Miraomics, we believe that data should be as useful as it is abundant. This release is just the beginning. We’re building the infrastructure and tools to ensure that every client can surface insights from their own experiments in the context of the world’s data. “We are excited to announce this major release of high quality curated data, representing thousands of hours of curation effort, enabling new opportunities for development of novel AI tools and novel insights in basic science, disease progression and drug discovery,” said Eugene Bolotin, Co-Founder and CEO at @Miraomics. https://lnkd.in/g7wgXDa4

I greatly enjoyed attending the QBI symposium, "Separating the Signal from Noise: AI in Biology." Many amazing talks were presented from basic science to drug discovery. Big thanks to the conference organizers and presenters. My key takeaway: AI in Biology is still very early, but amazingly promising. Most of the concepts involved "scientist in the loop", with AI serving as a virtual assistant, multiplying the speed the expert could do the work thousands of fold. However, we are still far from completely autonomous science. We are also thinking about accelerating research and we are grateful to the organizers for accepting our poster for presentation: "Towards an Easy Transformer-based Universal Cell-Typer". In it, we describe our latest efforts between: miraomics, RJ Honicky, Tùng Nguyễn from Pythia Biosciences to automate cell typing, also known as cell annotation, using transformer based methods. This is a foundational problem in biology, since identifying cells is often the first step in many biological experiments. Our framework is designed to be extendable across multi-omic datatypes such as: scRNA, proteomics, and image data through transformer enabled co-embedding. We are working on eliminating batch effects across methods and labs, benchmarking against current state-of-the-art. We aim to take the pain out of manual or semi manual cell typing, while keeping the "expert in the loop." If you are interested in learning more and getting the copy of the poster, getting early access to our code or collaborating, please don't hesitate to reach out! https://lnkd.in/gECiEH6h

We're looking forward to doing more great work with Latch!

We’re excited to present our latest progress as part of the collaboration between Pythia Biosciences and miraomics, focused on building cutting-edge methods for cell type annotation and multi-omic data interpretation. This joint effort brings together deep expertise in computational biology and single-cell analytics to unlock richer, more precise insights into complex biological systems.