[go: up one dir, main page]

Miraomics

Universal Cell
Classification

MiraTyper delivers fast, accurate, and reproducible cell classification across hundreds of cell types, multiple data modalities, and diverse tissues—without tissue-specific model selection.

377
Cell Types Classified

110M
Cells in 30 min

2-10×
Better Recall

Spatial Distribution: smooth muscle cell classification showing MiraTyper output

Cell Typing is Broken

Current approaches create significant bottlenecks that slow down research and introduce errors.

⏱️

Slow & Manual

Teams spend weeks hand-labeling cells. Manual pipelines are fragile, time-consuming, and error-prone.

🔀

Inconsistent Results

Public tools are slow, inconsistent across studies, and require tissue-specific model selection.

📉

Batch Effects

Accuracy and reproducibility suffer across different sequencing platforms and donors.

Better. Faster. Flexible.

A single universal model that works across tissues, platforms, and species.

High-Fidelity

Reproducible phenotyping across hundreds of cell types with hierarchical F1 scores of 0.84–0.99.

Lightning Fast

From hours to minutes. Process ~110M cells in 30 minutes. 30k cells in under 1 minute on CPU.

🔄

Universal Model

No manual model selection required. Works across tissues, platforms (10x, CosMx, Visium), and species.

MiraTyper vs. Current Methods

See how MiraTyper compares to traditional cell typing approaches.

Feature Current Methods MiraTyper
Approach Marker-based / Cluster-based Marker/Cluster Agnostic
Batch Effects Susceptible to batch variation Batch resistant
Platform Support Technology-specific (10x, CosMx) Cross-platform compatible
Tissue Scope Tissue-specific models required Cross-tissue with one model
Rare Cell Types Difficult to detect Excellent detection
Model Selection Many models / manual selection Single unified model
Label Harmonization Manual normalization required Pre-harmonized CL ontology

Validated Benchmarks

Independently validated against CellTypist using hierarchical F1 metrics on real-world datasets.

Blood (Immune)

Single-cell mapping of human ureter immune subset (Fink et al. 2022)

Hierarchical Precision

CellTypist: 0.87
MiraTyper: 0.87

Hierarchical Recall

CellTypist: 0.40
MiraTyper: 0.84

Hierarchical F1

CellTypist: 0.55
MiraTyper: 0.86

↑ 2× Higher Recall

Brain (Neocortex)

Transcriptomic cytoarchitecture of human neocortex (Jorstad 2023)

Hierarchical Precision

CellTypist: 0.71
MiraTyper: 0.99

Hierarchical Recall

CellTypist: 0.11
MiraTyper: 0.97

Hierarchical F1

CellTypist: 0.19
MiraTyper: 0.98

↑ 10× Higher Recall

Built on Cutting-Edge AI

MiraTyper combines hybrid generative AI embeddings with transformer-based classification.

🧬

100M+ Cell Training

Trained on CellXGene corpus with rigorous quality control

🏷️

CL Ontology Labels

Pre-harmonized Cell Ontology labels with confidence scores

🔬

Zero-Shot Learning

Handles rare and novel cell types without retraining

📊

Multi-Platform

10x Genomics, CosMx, Visium, and more

Accelerate Your Research

MiraTyper powers discovery and development across the drug and diagnostics pipeline.

💊

Drug Discovery

Identify detailed cell types and disease states in your proprietary or public data sets. Harmonize data sets across batches and technologies.

🔬

Pathology Model Training

Identify image patches with detailed cell mixtures using spatial transcriptomics data.

🩺

Diagnostic Development

Fine tune classifiers on novel conditions and utilize confidence metrics to identify atypical cell states.

🔍

Biomarker Discovery

Identify disease-associated cell states and rare populations with high precision.

Ready to Transform Your Cell Analysis?

Get started with MiraTyper today. Available as a Docker package or through our user-friendly application.