LAIOR Benchmarks
A comprehensive benchmarking library for single-cell analysis models, featuring LAIOR (Lorentz Attentive Interpretable ODE Regularized VAE)—the latest evolution in variational autoencoders from VAE → iVAE → LiVAE → LAIOR, integrating geometric regularization, information bottleneck, and ODE-based trajectory stabilization. Compare 23 models across 6 categories using 24 standardized metrics.
23
Models
0
RNA
0
ATAC
24
Metrics
Model Categories
Compare 6 distinct approaches in LAIOR's benchmarking suite, from contrastive learning to disentangled representations
Predictive Models
Encoder-only models without decoder reconstruction (contrastive learning)
Generative Models
VAE/autoencoder models with encoder + decoder reconstruction (general VAE class)
Trajectory Inference Models
VAE models specialized for developmental trajectory and pseudo-time inference
scATAC-Specific Models
VAE models specifically designed for chromatin accessibility data
Gaussian Geometric Models
VAE models using geometric structures (hyperbolic, mixture models, learnable curvature)
Disentanglement Models
VAE models focused on learning disentangled factor representations
Benchmark Datasets
LAIOR's curated collection of 0 single-cell datasets for rigorous model evaluation
Evaluation Metrics
24 standardized metrics across 4 evaluation categories for comprehensive model comparison
Clustering & Cell Type Discovery
Supervised metrics comparing predicted clusters to ground truth labels
Embedding Quality (UMAP & t-SNE)
Visualization quality via coranking analysis (4 metrics × 2 methods)
Intrinsic Latent Space (LSE)
Unsupervised geometric, spectral, and topological properties
Computational Efficiency
Training and inference performance
Platform Capabilities
Architecture Analysis
Detailed model comparisons including LAIOR's integrated architecture (ODE regularization + Lorentz geometry + information bottleneck + transformer attention) vs. LiVAE, iVAE, and classical VAE approaches
Multi-Modal Benchmarking
Evaluate models across 0 curated datasets spanning scRNA-seq (0) and scATAC-seq (0) modalities
Comprehensive Metrics
24 standardized metrics across clustering (6), embedding quality (8), latent space (8), and runtime (2)
Ready to explore?
Dive into LAIOR's benchmarking library and discover which models work best for your single-cell analysis