Skip to content

Examples Overview¤

Practical examples demonstrating DiffBio's differentiable bioinformatics operators, from single-operator basics to full pipeline composition with ecosystem integration.

Example Tiers¤

13 Basic
Single-operator patterns (5-10 min)
5 Intermediate
Multi-operator workflows (10-20 min)
14 Advanced
Pipelines & ecosystem integration (20-40 min)

Basic¤

Single-operator examples. One config, one apply(), one gradient check. Target audience: first-time DiffBio users.

Example Duration Key Operators Description
Operator Pattern 5 min SoftKMeansClustering The universal Config -> Construct -> Apply pattern
MolNet Data Loading 5 min MolNetSource Load MoleculeNet benchmark datasets
Molecular Fingerprints 10 min CircularFingerprint Generate ECFP and neural fingerprints
Molecular Similarity 5 min TanimotoSimilarity Compare molecules with similarity metrics
Scaffold Splitting 5 min ScaffoldSplitter Proper train/test splits for drug discovery
DNA Encoding 5 min one-hot encoding One-hot encode DNA sequences
Sequence Alignment 10 min SmoothSmithWaterman Smith-Waterman local alignment
Pileup Generation 10 min DifferentiablePileup Generate pileups from aligned reads
Single-Cell Clustering 10 min SoftKMeansClustering Soft k-means with training loop
RNA Structure 10 min McCaskill algorithm Predict RNA secondary structure
Protein Structure 10 min DSSP prediction Predict protein secondary structure
HMM Sequence Model 10 min DifferentiableHMM Hidden Markov Models for sequences
Preprocessing 10 min QualityFilter, AdapterRemoval Read preprocessing pipeline

Intermediate¤

Multi-operator workflows with two or three operators chained, parameter sweeps, or evaluation against ground truth. Target audience: users building custom pipelines.

Example Duration Key Operators Description
Imputation 15 min DifferentiableDiffusionImputer MAGIC-style diffusion imputation for dropout recovery
Trajectory 20 min DifferentiablePseudotime, FateProbability, SwitchDE Pseudotime ordering and fate probability estimation
Cell Annotation 15 min DifferentiableCellAnnotator (3 modes) Cell type annotation: celltypist, cellassign, scanvi
Doublet Detection 15 min DoubletScorer, SoloDetector Scrublet-style and Solo-style doublet detection
Batch Correction 20 min Harmony, MMD, WGAN Three batch correction strategies compared

Advanced¤

Full pipeline composition with ecosystem integration (calibrax metrics, artifex losses, opifex training). Training loops, benchmarking, and multi-operator chains. Target audience: researchers adapting DiffBio for their data.

Example Duration Key Operators Description
Spatial Analysis 25 min SpatialDomain, PASTEAlignment STAGATE domain identification and PASTE slice alignment
GRN Inference 25 min DifferentiableGRN Gene regulatory network inference via GATv2 attention
Single-Cell Pipeline 30 min Simulator, AmbientRemoval, Imputer, Clustering, Pseudotime Five-operator end-to-end pipeline
Calibrax Metrics 25 min SoftKMeansClustering, DifferentiableAUROC Training vs evaluation metric split with calibrax
scVI Benchmark 30 min VAENormalizer, MultiOmicsVAE scVI-style VAE training with calibrax evaluation
Drug Discovery Workflow 30 min CircularFingerprint, PropertyPredictor End-to-end drug discovery pipeline
ADMET Prediction 25 min ADMETPredictor Multi-task ADMET property prediction
AttentiveFP GNN 25 min AttentiveFPOperator Attention-based molecular fingerprints
Variant Calling Pipeline 30 min Full variant calling pipeline End-to-end variant calling with CNN classifier
Single-Cell Batch Correction 20 min DifferentiableHarmony Harmony-style batch correction
Differential Expression 25 min NB-GLM DESeq2-style statistical testing
RNA Velocity 25 min Neural ODE velocity RNA velocity trajectory inference
Epigenomics Analysis 25 min Peak calling, chromatin states ChIP-seq and ATAC-seq analysis
Multi-omics Integration 30 min Spatial deconvolution, Hi-C Multi-omics data integration

Running Examples¤

All examples are self-contained Python scripts that generate synthetic data and produce verifiable outputs.

# Setup
./setup.sh
source ./activate.sh

# Run any example
uv run python examples/basics/operator_pattern.py
uv run python examples/singlecell/clustering.py
uv run python examples/ecosystem/scvi_benchmark.py

Key Features Demonstrated¤

All examples showcase DiffBio's core capabilities:

  1. Differentiability -- every operator supports jax.grad for gradient computation
  2. JIT Compilation -- all operators work with jax.jit for accelerated execution
  3. apply() Contract -- consistent result, state, metadata = operator.apply(data, {}, None) interface
  4. Synthetic Data -- self-contained examples with no external data dependencies
  5. Ecosystem Integration -- calibrax metrics, artifex losses, and opifex training utilities

Contributing Examples¤

See the Contributing Guide and the Example Documentation Design Guide for details on adding new examples.