bio-clip-seq-clip-deep-learning

What this skill does

## Version Compatibility

Reference examples tested with: RBPNet (Jens & Gagneur 2024 github), RNAProt 0.5+, GraphProt2 (Sauer 2022 github), DeepCLIP 1.0+ (Gronning 2020), DeepRiPe (Krakau lab), pytorch 2.2+, tensorflow 2.15+, scikit-learn 1.4+, biopython 1.83+, transformers 4.40+ (for RNA foundation models).

Before using code patterns, verify installed versions match. If versions differ:
- Python: `pip show <package>` then `help(module.function)` to check signatures
- Frameworks: check pytorch / tensorflow versions; reproducibility depends on framework version

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

# CLIP-seq Deep Learning

**"Predict RBP binding from RNA sequence using deep learning"** -> Train or apply neural networks that learn the sequence (and optionally structure) preference of an RBP from CLIP-seq peaks or single-nucleotide crosslink sites. The output is per-base or per-site binding probability for any input sequence, enabling: (a) variant-effect prediction at heterozygous SNPs; (b) in silico binding-site discovery on transcripts not covered by CLIP; (c) systematic comparison across RBPs via shared model architectures; (d) interpretation via attribution / saliency to recover RBP-specific motifs and structural preferences. Modern models (RBPNet 2024) predict per-nucleotide crosslink count distributions rather than binary peak/non-peak, providing single-nt resolution outputs.

- Python (RBPNet sequence-to-CL signal): `import rbpnet; model = rbpnet.load_pretrained('RBP_name'); predictions = model.predict(sequence)` produces per-base CL count distribution
- Python (RNAProt RNN classifier): `RNAProt train -i peaks.bed -t background.bed -g genome.fa -o model/` then `RNAProt predict -m model/ -i query_sequences.fa -o predictions.tsv`
- Python (GraphProt2 GCN with structure): `graphprot2 train -i peaks.bed -bg shuffled.bed -g genome.fa --structure -o model/`
- Python (DeepCLIP for binding probability): `deepclip --train --train_data train.fa --validation_data val.fa --predict --predict_data test.fa --output_dir output/`
- Python (DeepRiPe multi-modal CNN): `from deepripe import DeepRiPe; model.train(X_train, y_train); predictions = model.predict(X_test)`

The benchmarks (RNAProt paper, 2021): RNAProt AUC 87-89%; DeepCLIP 84-87%; GraphProt 82-84%. RBPNet (2024) is the modern sequence-to-signal model that predicts per-nt CL distributions at single-nucleotide resolution rather than binary site classification.

## Models Taxonomy

| Model | Architecture | Input | Output | Resolution | Strength | Fails when |
|-------|--------------|-------|--------|------------|----------|------------|
| RBPNet (Jens & Gagneur 2024) | Sequence-to-signal CNN | Sequence | Per-nt CL count distribution | Single-nt | Modern single-nt resolution; predicts CL distribution not binary | New (2024); fewer pretrained RBPs |
| RNAProt (Uhl 2021) | LSTM RNN | Sequence | Binary binding probability | Site/peak | Highest AUC in benchmark (87-89%); feature-rich (RNA-Foldscan structure) | Single-prediction; not per-base profile |
| GraphProt (Maticzka 2014) | Graph kernel + SVM | Sequence + structure | Binary | Site/peak | Original structure-aware; well-validated | Older; superseded by GraphProt2 |
| GraphProt2 (Sauer 2022) | Graph Convolutional Network (GCN) | Variable-length sequence + structure | Nucleotide-wise binding profile | Per-nt | Variable-length input; structure-aware | Slow to train; needs GPU |
| DeepCLIP (Gronning 2020) | CNN | Sequence | Binary | Site | Fast; good benchmarks | Sequence-only; no structure |
| DeepRiPe (Ghanbari 2020) | Multi-modal CNN | Sequence + structure + region | Binary | Site | Multi-input; good ENCODE benchmark | Sequence/structure/region must be pre-extracted |
| iDeep / iDeepE (Pan 2018) | CNN ensemble | Sequence | Binary | Site | Ensemble approach | Older; few pretrained models |
| Pysster (Budach 2018) | CNN-LSTM | Sequence | Binary | Site | Generic framework | Less RBP-specific |
| DeepBind (Alipanahi 2015) | CNN | Sequence | Binary | Site | First deep-learning RBP model | Outdated; superseded |
| BasenjiPredict (Yeo lab adapt) | Multi-task CNN | Sequence | Per-task profile | Per-nt | Joint learning across RBPs | Computational overhead |
| RNA foundation models (RNAErnie, RNA-FM 2024) | Transformer pretrained on RNA | Sequence | Embeddings (downstream task) | Embedding | Transfer learning across RBPs | Foundation model trained at depth; fine-tuning needed |

Methodology evolves; verify the latest publication on RBP deep learning (RBPNet 2024 is the current state-of-the-art per-nt resolution model). RNA foundation models (RNA-FM, RNAErnie) are emerging in 2024 as transfer-learning backbones; fine-tuning on CLIP data for specific RBPs is the next-generation approach.

## Critical Choice: Binary Classification vs Sequence-to-Signal

**Binary classification (RNAProt, DeepCLIP, DeepRiPe, GraphProt2):** Train on labeled site vs background; predict probability of binding for an input sequence. Output: per-sequence score. Pro: simple framework; mature benchmarks. Con: discards single-nt CL distribution information; binary decision boundary.

**Sequence-to-signal (RBPNet):** Train on per-nt CL count distributions from PureCLIP or CTK CITS output; predict per-base CL count for input sequence. Output: per-nt profile. Pro: single-nt resolution; preserves CL count information; matches biology (CL is a sharp signal, not a region). Con: newer (2024); fewer pretrained models; harder to interpret with classical motif tools.

| Goal | Model |
|------|-------|
| Predict RBP binding probability for an input sequence | RNAProt or DeepRiPe |
| Predict per-base CL distribution | RBPNet |
| Variant-effect at heterozygous SNP | RBPNet or DeepRiPe (per-base output) |
| Compare RBP preferences across ENCODE | Multi-task model (BasenjiPredict-style) |
| Transfer learning across RBPs | RNA foundation model + fine-tune |
| In silico screening of variants | RBPNet (per-base) for genome-wide |
| Motif interpretation via attribution | DeepRiPe or GraphProt2 (interpretable) |
| Custom training on new CLIP data | RNAProt (easiest pipeline) |
| Production-grade per-base prediction | RBPNet 2024 |

## Variant-Effect Prediction Workflow

Apply a pretrained or custom-trained model to predict the change in binding upon a sequence variant.

```python
import torch
from rbpnet import RBPNet  # hypothetical API; verify per-package documentation

# Load pretrained model for specific RBP
model = RBPNet.load_pretrained('TARDBP_HEK293T')

# Reference and alternative sequences around a variant
ref_seq = 'CTGTACTGCAGTAGCATGCTAGCATGCTAGCAT'  # 32 nt window centered on variant
alt_seq = 'CTGTACTGCAGTAGCATGCTAGCATGCTAGCAA'  # Variant: T -> A at position 32

# Predict per-base CL distribution for both
ref_pred = model.predict(ref_seq)   # shape: (32, 1) - per-base CL probability
alt_pred = model.predict(alt_seq)

# Variant effect: log2 fold change in summed binding signal
import numpy as np
effect = np.log2((alt_pred.sum() + 1e-9) / (ref_pred.sum() + 1e-9))
print(f'Variant effect (log2 FC): {effect:.4f}')

# Strong-effect variant: |log2 FC| > 1.0
# Mid-effect: 0.5 - 1.0
# Weak: < 0.5
```

For genome-wide variant scoring: apply this in batch to all GWAS variants overlapping the RBP's binding regions. The output is a per-variant log2 FC; downstream Mendelian randomization or fine-mapping integrates with phenotype-association statistics.

## Training a Custom Model (RNAProt Example)

RNAProt is the most accessible training framework. It accepts peak BED + background BED + genome FASTA.

**Goal:** Train a chromosome-split RNN classifier from CLIP peaks to predict RBP binding probability on arbitrary input sequences, with held-out evaluation on a chromosome-distinct test set.

**Approach:** Generate a GC-matched 3' UTR background, split peaks and background by chromosome (train c