---
license: apache-2.0
---

![Overview of PeptiVerse](peptiverse-cover.png)

# PeptiVerse: A Unified Platform for Therapeutic Peptide Property Prediction 🧬🌌

This is the repository for [PeptiVerse: A Unified Platform for Therapeutic Peptide Property Prediction](https://www.biorxiv.org/content/10.64898/2025.12.31.697180), a collection of machine learning predictors for canonical and non-canonical peptide property prediction using sequence and SMILES representations. 🧬 PeptiVerse 🌌 enables evaluation of key biophysical and therapeutic properties of peptides for property-optimized generation.

## Table of Contents 🌟

- [Quick start](#quick-start)
- [Installation](#installation)
- [Repository Structure](#repository-structure)
- [Training data collection](#training-data-collection)
- [Best model list](#best-model-list)
   - [Full model set (cuML-enabled)](#full-model-set-gpu-enabled)
   - [Minimal deployable model set (no cuML)](#minimal-deployable-set)
- [Usage](#usage)
   - [Local Application Hosting](#local-application-hosting)
   - [Dataset integration](#dataset-integration)
   - [Training](#training)
   - [Quick inference by property per model](#Quick-inference-by-property-per-model)
- [Property Interpretations](#property-interpretations)
- [Model Architecture](#model-architecture)
- [Troubleshooting](#troubleshooting)
- [Citation](#citation)

## Quick Start 🌟
- Light-weighted start (basic models, no cuML, read below for details)
```bash
# Ignore all LFS files, you will see an empty folder first
git clone --no-checkout https://huggingface.co/ChatterjeeLab/PeptiVerse
cd PeptiVerse

# Enable sparse checkout
git sparse-checkout init --cone

# Choose only selective items to download
git sparse-checkout set \
  inference.py \
  download_light.py \
  best_models.txt \
  basic_models.txt \
  requirements.txt \
  tokenizer \
  README.md

# Now checkout
GIT_LFS_SKIP_SMUDGE=1 git checkout

# Install basic pkgs
pip install -r requirements.txt

# Download basic model weights according to the basic_models.txt. Adjust which config you wanted as needed.
python download_light.py

# Test in inference
python inference.py
```
- Full model clone (will clone all model weights)
```bash
# Clone repository
git clone https://huggingface.co/ChatterjeeLab/PeptiVerse

# Install dependencies
pip install -r requirements.txt

# Run inference
python inference.py
```
## Installation 🌟
### Minimal Setup
- Easy start-up environment (using transformers, xgboost models)
```bash
pip install -r requirements.txt
```
### Full Setup
- Additional access to trained SVM and ElastNet models requires installation of `RAPIDS cuML`, with instructions available from their official [github page](https://github.com/rapidsai/cuml) (**CUDA-capable GPU required**).
- Optional: pre-compiled Singularity/Apptainer environment (7.52G) is available at [Google drive](https://drive.google.com/file/d/1RJQ9HK0_gsPOhRo5H5ZmH_MYcpJqQD7e/view?usp=sharing) with everything you need (still need CUDA/GPU to load cuML models).
    ```
    # test
    apptainer exec peptiverse.sif python -c "import sys; print(sys.executable)"

    # run inference (see below)
    apptainer exec peptiverse.sif python inference.py
    ```
## Repository Structure 🌟
This repo contains important large files for [PeptiVerse](https://huggingface.co/spaces/ChatterjeeLab/PeptiVerse), an interactive app for peptide property prediction. [Paper link.](https://www.biorxiv.org/content/10.64898/2025.12.31.697180v1)

```
PeptiVerse/
├── training_data_cleaned/     # Processed datasets with embeddings
│   └── <property>/            # Property-specific data
│       ├── train/val splits
│       └── precomputed embeddings
├── training_classifiers/      # Trained model weights
│   └── <property>/           
│       ├── cnn_wt/           # CNN architectures
│       ├── mlp_wt/           # MLP architectures
│       └── xgb_wt/           # XGBoost models
├── tokenizer/                 # PeptideCLM tokenizer
├── training_data/             # Raw training data
├── inference.py               # Main prediction interface
├── best_models.txt            # Model selection manifest
└── requirements.txt           # Python dependencies
```
For full data access, please download the corresponding `training_data_cleaned` and `training_classifiers` from zenodo. The current Huggingface repo only hosts best model weights and meta data with splits labels.

## Training Data Collection 🌟

<table>
  <caption><strong>Data distribution.</strong> Classification tasks report counts for class 0/1; regression tasks report total sample size (N).</caption>
  <thead>
    <tr>
      <th rowspan="2"><strong>Properties</strong></th>
      <th colspan="2"><strong>Amino Acid Sequences</strong></th>
      <th colspan="2"><strong>SMILES Sequences</strong></th>
    </tr>
    <tr>
      <th><strong>0</strong></th>
      <th><strong>1</strong></th>
      <th><strong>0</strong></th>
      <th><strong>1</strong></th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td colspan="5"><strong>Classification</strong></td>
    </tr>
    <tr>
      <td>Hemolysis</td>
      <td>4765</td>
      <td>1311</td>
      <td>4765</td>
      <td>1311</td>
    </tr>
    <tr>
      <td>Non-Fouling</td>
      <td>13580</td>
      <td>3600</td>
      <td>13580</td>
      <td>3600</td>
    </tr>
    <tr>
      <td>Solubility</td>
      <td>9668</td>
      <td>8785</td>
      <td>9668</td>
      <td>8785</td>
    </tr>
    <tr>
      <td>Permeability (Penetrance)</td>
      <td>1162</td>
      <td>1162</td>
      <td>1162</td>
      <td>1162</td>
    </tr>
    <tr>
      <td>Toxicity</td>
      <td>-</td>
      <td>-</td>
      <td>5518</td>
      <td>5518</td>
    </tr>
    <tr>
      <td colspan="5"><strong>Regression (N)</strong></td>
    </tr>
    <tr>
      <td>Permeability (PAMPA)</td>
      <td colspan="2" align="center">-</td>
      <td colspan="2" align="center">6869</td>
    </tr>
    <tr>
      <td>Permeability (CACO2)</td>
      <td colspan="2" align="center">-</td>
      <td colspan="2" align="center">606</td>
    </tr>
    <tr>
      <td>Half-Life</td>
      <td colspan="2" align="center">130</td>
      <td colspan="2" align="center">245</td>
    </tr>
    <tr>
      <td>Binding Affinity</td>
      <td colspan="2" align="center">1436</td>
      <td colspan="2" align="center">1597</td>
    </tr>
  </tbody>
</table>


## Best Model List 🌟

### Full model set (cuML-enabled)
| Property | Best Model (Sequence) | Best Model (SMILES) | Task Type | Threshold (Sequence) | Threshold (SMILES) |
|---|---|---|---|---|---|
| Hemolysis | SVM | CNN (chemberta) | Classifier | 0.2521 | 0.564 |
| Non-Fouling | Transformer | ENET (peptideclm) | Classifier | 0.57 | 0.6969 |
| Solubility | CNN | – | Classifier | 0.377 | – |
| Permeability (Penetrance) | SVM | SVM (chemberta) | Classifier | 0.5493 | 0.573 |
| Toxicity | – | CNN (chemberta) | Classifier | – | 0.49 |
| Binding Affinity | unpooled | unpooled | Regression | – | – |
| Permeability (PAMPA) | – | CNN (chemberta) | Regression | – | – |
| Permeability (Caco-2) | – | SVR (chemberta) | Regression | – | – |
| Half-life | Transformer | XGB (peptideclm) | Regression | – | – |

>Note: *unpooled* indicates models operating on token-level embeddings with cross-attention, rather than mean-pooled representations.

### Minimal deployable model set (no cuML)
| Property | Best Model (WT) | Best Model (SMILES) | Task Type | Threshold (WT) | Threshold (SMILES) |
|---|---|---|---|---|---|
| Hemolysis | XGB | CNN (chemberta) | Classifier | 0.2801 | 0.564 |
| Non-Fouling | Transformer | XGB (peptideclm) | Classifier | 0.57 | 0.3892 |
| Solubility | CNN | – | Classifier | 0.377 | – |
| Permeability (Penetrance) | XGB | XGB (chemberta) | Classifier | 0.4301 | 0.5028 |
| Toxicity | – | CNN (chemberta) | Classifier | – | 0.49 |
| Binding Affinity | wt_wt_pooled | chemberta_smiles_pooled | Regression | – | – |
| Permeability (PAMPA) | – | CNN (chemberta) | Regression | – | – |
| Permeability (Caco-2) | – | SVR (chemberta) | Regression | – | – |
| Half-life | Transformer | XGB (peptideclm) | Regression | – | – |
>Note: Models marked as SVM or ENET are replaced with XGB as these models are not currently supported in the deployment environment without cuML setups.


## Usage 🌟

### Local Application Hosting
- Host the [PeptiVerse UI](https://huggingface.co/spaces/ChatterjeeLab/PeptiVerse) locally with your own resources. 
```bash
# Configure models in best_models.txt

git clone https://huggingface.co/spaces/ChatterjeeLab/PeptiVerse
python app.py
```
### Data pre-processing
Under the `training_data_cleaned`, we provided the generated embeddings in huggingface dataset format. The following scripts are the steps used to generate the data.

### Dataset integration
- All properties are provided with raw_data/split_ready_csvs/[huggingface_datasets](https://huggingface.co/docs/datasets/en/index).
- Selective download the data you need with `huggingface-cli`
```bash
huggingface-cli download ChatterjeeLab/PeptiVerse \
  --include "training_data_cleaned/**" \     # only this folder
  --exclude "**/*.pt" "**/*.joblib" \     # skip weights/artifacts
  --local-dir PeptiVerse_partial \
  --local-dir-use-symlinks False      # make real copies
```
- Or in python
```python
from huggingface_hub import snapshot_download

local_dir = snapshot_download(
    repo_id="ChatterjeeLab/PeptiVerse",
    allow_patterns=["training_data_cleaned/**"],     # only this folder
    ignore_patterns=["**/*.pt", "**/*.joblib"],     # skip weights/artifacts
    local_dir="PeptiVerse_partial",
    local_dir_use_symlinks=False,                   # make real copies
)
print("Downloaded to:", local_dir)
```
- Usage of the huggingface datasets (with pre-computed embeddings and splits)
    - All embedding datasets are saved via `DatasetDict.save_to_disk` and loadable with:
    ``` python
    from datasets import load_from_disk
    ds = load_from_disk(PATH)
    train_ds = ds["train"]
    val_ds = ds["val"]
    ```
- A) Sequence Based ([ESM-2](https://huggingface.co/facebook/esm2_t33_650M_UR50D) embeddings)
    - Pooled (fixed-length vector per sequence)
        - Generated by mean-pooling token embeddings excluding special tokens (CLS/EOS) and padding.
        - Each item:
            sequence: `str`
            label: `int` (classification) or `float` (regression)
            embedding: `float32[H]` (H=1280 for ESM-2 650M)
    - Unpooled (variable-length token matrix)
        - Generated by keeping all valid token embeddings (excluding special tokens + padding) as a per-sequence matrix.
        - Each item:
            sequence: `str`
            label: `int` (classification) or `float` (regression)
            embedding: `float16[L, H]` (nested lists)
            attention_mask: `int8[L]`
            length: `int` (=L)
- B) SMILES-based ([PeptideCLM](https://github.com/AaronFeller/PeptideCLM) embeddings)
    - Pooled (fixed-length vector per sequence)
        - Generated by mean-pooling token embeddings excluding special tokens (CLS/EOS) and padding.
        - Each item:
            sequence: `str` (SMILES)
            label: `int` (classification) or `float` (regression)
            embedding: `float32[H]`
    - Unpooled (variable-length token matrix)
        - Generated by keeping all valid token embeddings (excluding special tokens + padding) as a per-sequence matrix.
        - Each item:
            sequence: `str` (SMILES)
            label: `int` (classification) or `float` (regression)
            embedding: `float16[L, H]` (nested lists)
            attention_mask: `int8[L]`
            length: `int` (=L)
- C) SMILES-based ([ChemBERTa](https://huggingface.co/DeepChem/ChemBERTa-77M-MLM) embeddings)
    - Pooled (fixed-length vector per sequence)
        - Generated by mean-pooling token embeddings excluding special tokens (CLS/EOS) and padding.
        - Each item:
            sequence: `str` (SMILES)
            label: `int` (classification) or `float` (regression)
            embedding: `float32[H]`
    - Unpooled (variable-length token matrix)
        - Generated by keeping all valid token embeddings (excluding special tokens + padding) as a per-sequence matrix.
        - Each item:
            sequence: `str` (SMILES)
            label: `int` (classification) or `float` (regression)
            embedding: `float16[L, H]` (nested lists)
            attention_mask: `int8[L]`
            length: `int` (=L)
### Training
Under the `training_classifiers` folder, we provide the python scripts used to train different models. The scripts will 
1. Read the pre-processed Huggingface Dataset from `training_data_cleaned` folder; 
2. Perform OPTUNA hyperparameter sweep once being called; 
3. All training was conducted on HPC with SLURM script under `training_classifiers/src` folder; 
4. Customize or isolate certain model training scripts as needed.
##### Example of training
###### ML models
```
HOME_LOC=/home
SCRIPT_LOC=$HOME_LOC/PeptiVerse/training_classifiers
EMB_LOC=$HOME_LOC/PeptiVerse/training_data_cleaned

OBJECTIVE='hemolysis' # nf/solubility/hemolysis/permeability_pampa/permeability_caco2
WT='smiles' # wt/smiles
DATA_FILE="hemo_${WT}_with_embeddings"
LOG_LOC=$SCRIPT_LOC/src/logs
DATE=$(date +%m_%d)
MODEL_TYPE='svm_gpu' # xgb/enet_gpu/svm_gpu
SPECIAL_PREFIX="${MODEL_TYPE}-${OBJECTIVE}-${WT}_new"

# Create log directory if it doesn't exist
mkdir -p $LOG_LOC

cd $SCRIPT_LOC

python -u train_ml.py \
  --dataset_path "${DATA_LOC}/${OBJECTIVE}/${DATA_FILE}" \
  --out_dir "${SCRIPT_LOC}/${OBJECTIVE}/${MODEL_TYPE}_${WT}" \
  --model "${MODEL_TYPE}" \
  --n_trials 200  > "${LOG_LOC}/${DATE}_${SPECIAL_PREFIX}.log" 2>&1
```
###### DNN models
```
HOME_LOC=/home
SCRIPT_LOC=$HOME_LOC/PeptiVerse/training_classifiers
EMB_LOC=$HOME_LOC/PeptiVerse/training_data_cleaned

OBJECTIVE='nf' # nf/solubility/hemolysis
WT='smiles' #wt/smiles
DATA_FILE="nf_${WT}_with_embeddings_unpooled"
LOG_LOC=$SCRIPT_LOC/src/logs
DATE=$(date +%m_%d)
MODEL_TYPE='cnn' #mlp/cnn/transformer
SPECIAL_PREFIX="${MODEL_TYPE}-${OBJECTIVE}-${WT}"

# Create log directory if it doesn't exist
mkdir -p $LOG_LOC

cd $SCRIPT_LOC

python -u train_nn.py \
  --dataset_path "${DATA_LOC}/${OBJECTIVE}/${DATA_FILE}" \
  --out_dir "${SCRIPT_LOC}/${OBJECTIVE}/${MODEL_TYPE}_${WT}" \
  --model "${MODEL_TYPE}" \
  --n_trials 200  > "${LOG_LOC}/${DATE}_${SPECIAL_PREFIX}.log" 2>&1
```
###### Binding Affinity
```
HOME_LOC=/home
SCRIPT_LOC=$HOME_LOC/PeptiVerse/training_classifiers
EMB_LOC=$HOME_LOC/PeptiVerse/training_data_cleaned

OBJECTIVE='binding_affinity'
BINDER_MODEL='chemberta'   # peptideclm / chemberta
STATUS='unpooled'             # pooled / unpooled
TYPE='smiles'
DATA_FILE='pair_wt_${TYPE}_${STATUS}'

LOG_LOC=$SCRIPT_LOC/src/logs
DATE=$(date +%m_%d)
SPECIAL_PREFIX="${OBJECTIVE}-${BINDER_MODEL}-${STATUS}"

python -u binding_training.py \
  --dataset_path "${EMB_LOC}/${OBJECTIVE}/${BINDER_MODEL}/${DATA_FILE}" \
  --mode "${STATUS}" \
  --out_dir "${SCRIPT_LOC}/${OBJECTIVE}/${BINDER_MODEL}_${TYPE}_${STATUS}" \
  --n_trials 200 > "${LOG_LOC}/${DATE}_${SPECIAL_PREFIX}.log" 2>&1
```

### Quick inference by property per model
```python
from inference import PeptiVersePredictor
from pathlib import Path

root = Path(__file__).resolve().parent  # current script folder

    
predictor = PeptiVersePredictor(
    manifest_path=root / "best_models.txt",
    classifier_weight_root=root,
    device="cuda",                            # or "cpu"
)

# mode: smiles (SMILES-based models) / wt (Sequence-based models) 
# property keys (with some level of name normalization)
# hemolysis
# nf (Non-Fouling)
# solubility
# permeability_penetrance
# toxicity
# permeability_pampa
# permeability_caco2
# halflife
# binding_affinity

seq = "GIVEQCCTSICSLYQLENYCN"
smiles = "CC(C)C[C@@H]1NC(=O)[C@@H](CC(C)C)N(C)C(=O)[C@@H](C)N(C)C(=O)[C@H](Cc2ccccc2)NC(=O)[C@H](CC(C)C)N(C)C(=O)[C@H]2CCCN2C1=O"

# Hemolysis
out = pred.predict_property("hemolysis", mode="wt", input_str=seq)
print(out)
# {"property":"hemolysis","mode":"wt","score":prob,"label":0/1,"threshold":...}

out = pred.predict_property("hemolysis", mode="smiles", input_str=smiles)
print(out)

# Non-fouling (key is nf)
out = pred.predict_property("nf", mode="wt", input_str=seq)
print(out)

out = pred.predict_property("nf", mode="smiles", input_str=smiles)
print(out)

# Solubility (Sequence-only)
out = pred.predict_property("solubility", mode="wt", input_str=seq)
print(out)

# Permeability (Penetrance) (Sequence-only)
out = pred.predict_property("permeability_penetrance", mode="wt", input_str=seq)
print(out)

# Toxicity (SMILES-only)
out = pred.predict_property("toxicity", mode="smiles", input_str=smiles)
print(out)

# Permeability (PAMPA) (SMILES regression)
out = pred.predict_property("permeability_pampa", mode="smiles", input_str=smiles)
print(out)
# {"property":"permeability_pampa","mode":"smiles","score":value}

# Permeability (Caco-2) (SMILES regression)
out = pred.predict_property("permeability_caco2", mode="smiles", input_str=smiles)
print(out)

# Half-life (sequence-based + SMILES regression)
out = pred.predict_property("halflife", mode="wt", input_str=seq)
print(out)

out = pred.predict_property("halflife", mode="smiles", input_str=smiles)
print(out)

# Binding Affinity
protein = "MTEYKLVVVGAGGVGKSALTIQLIQNHFVDEYDPTIEDSYRKQV..."  # target protein
peptide_seq = "GIVEQCCTSICSLYQLENYCN"

out = pred.predict_binding_affinity(
    mode="wt",
    target_seq=protein,
    binder_str=peptide_seq,
)
print(out)
# {
#   "property":"binding_affinity",
#   "mode":"wt",
#   "affinity": float,
#   "class_by_threshold": "High (≥9)" / "Moderate (7-9)" / "Low (<7)",
#   "class_by_logits": same buckets,
#   "binding_model": "pooled" or "unpooled",
# }

```

#### Advanced inference with uncertainty prediction
The uncertainty prediction is added as a parameter in the inference code. The full classifier folder from [zenodo]() is required to enable this functionality. The model uncertainty is reported via all the scripts listed under the `training_classifiers` folder starting with "**refit**". Detailed description can be found in the methodology part of the manuscript.
At inference time, PeptiVersePredictor returns an `uncertainty` field with every prediction when `uncertainty=True` is passed. The method and interpretation depend on the model class, determined automatically at inference time.
```python
seq = "GIGAVLKVLTTGLPALISWIKRKRQQ"
smiles = "C(C)C[C@@H]1NC(=O)[C@@H]2CCCN2C(=O)[C@@H](CC(C)C)NC(=O)[C@@H](CC(C)C)N(C)C(=O)[C@H](C)NC(=O)[C@H](Cc2ccccc2)NC1=O"

print(predictor.predict_property("nf",    "wt",     seq, uncertainty=True))
print(predictor.predict_property("nf",    "smiles",     smiles, uncertainty=True))

{'property': 'nf', 'col': 'wt', 'score': 0.00014520535252195523, 'emb_tag': 'wt', 'label': 0, 'threshold': 0.57, 'uncertainty': 0.0017192508727321288, 'uncertainty_type': 'ensemble_predictive_entropy'}
{'property': 'nf', 'col': 'smiles', 'score': 0.025485480204224586, 'emb_tag': 'peptideclm', 'label': 0, 'threshold': 0.6969, 'uncertainty': 0.11868063130587676, 'uncertainty_type': 'binary_predictive_entropy_single_model'}
```
 
---
 
##### Method by Model Class
 
| Model Class | Task | Uncertainty Method | Output Type | Range |
|---|---|---|---|---|
| MLP, CNN, Transformer | Classifier | Deep ensemble predictive entropy (5 seeds) | `float` | [0, ln(2) ≈ 0.693] |
| MLP, CNN, Transformer | Regression | Adaptive conformal interval; falls back to ensemble std if no MAPIE bundle | `(lo, hi)` or `float` | unbounded |
| SVM / SVC / XGBoost | Classifier | Binary predictive entropy (sigmoid of decision function) | `float` | [0, ln(2) ≈ 0.693] |
| SVR / ElasticNet / XGBoost | Regression | Adaptive conformal interval | `(lo, hi)` | unbounded |
 
> **Uncertainty is `None`** when: a DNN classifier has no seed ensemble trained, or a regression model has no `mapie_calibration.joblib` in its model directory.
 
---
## Interpretation 🌟 

You can also find the same description in the paper or in the PeptiVerse app `Documentation` tab.

---

### 🩸 Hemolysis Prediction<br>
50% of read blood cells being lysed at x ug/ml concetration (HC50). If HC50 < 100uM, considered as hemolytic, otherwise non-hemolytic, resulting in a binary 0/1 dataset. The predicted probability should therefore be interpreted as a risk indicator, not an exact concentration estimate.<br>

**Output interpretation:**<br>

- Score close to 1.0 = high probability of red blood cell membrane disruption
- Score close to 0.0 = non-hemolytic

---

### 💧 Solubility Prediction<br>
Outputs a probability (0–1) that a peptide remains soluble in aqueous conditions.<br>

**Output interpretation:**<br>

- Score close to 1.0 = highly soluble
- Score close to 0.0 = poorly soluble

---

### 👯 Non-Fouling Prediction<br>
Higher scores indicate stronger non-fouling behavior, desirable for circulation and surface-exposed applications.<br>

**Output interpretation:**<br>

- Score close to 1.0 = non-fouling
- Score close to 0.0 = fouling
  
---

### 🪣 Permeability Prediction<br>
Predicts membrane permeability on a log P scale.<br>

**Output interpretation:**<br>

- Higher values = more permeable (>-6.0)
- For penetrance predictions, it is a classification prediction, so within the [0, 1] range, closer to 1 indicates more permeable.

---

### ⏱️ Half-Life Prediction<br>
**Interpretation:** Predicted values reflect relative peptide stability for the unit in hours. Higher scores indicate longer persistence in serum, while lower scores suggest faster degradation.<br>

---

### ☠️ Toxicity Prediction<br>
**Interpretation:** Outputs a probability (0–1) that a peptide exhibits toxic effects. Higher scores indicate increased toxicity risk.<br>

---

### 🔗 Binding Affinity Prediction <br>

Predicts peptide-protein binding affinity. Requires both peptide and target protein sequence.<br>

**Interpretation:**<br>

- Scores ≥ 9 correspond to tight binders (K ≤ 10⁻⁹ M, nanomolar to picomolar range)<br>
- Scores between 7 and 9 correspond to medium binders (10⁻⁷–10⁻⁹ M, nanomolar to micromolar range)<br>
- Scores < 7 correspond to weak binders (K ≥ 10⁻⁶ M, micromolar and weaker)<br>
- A difference of 1 unit in score corresponds to an approximately tenfold change in binding affinity.<br>
    
---

### Uncertainty Interpretation <br>
#### Entropy (classifiers)<br>
 
Binary predictive entropy of the output probability p̄:<br>
 
$$\mathcal{H} = -\bar{p}\log\bar{p} - (1 - \bar{p})\log(1 - \bar{p})$$<br>
 
- For **DNN classifiers**: p̄ is the mean probability across 5 independently seeded models (deep ensemble). High entropy reflects both epistemic uncertainty (seed disagreement) and aleatoric uncertainty (collectively diffuse predictions).<br>

- For **XGBoost / SVM / ElasticNet classifiers**: p̄ is the single model's output probability (or sigmoid of decision function for ElasticNet). Entropy reflects output confidence of a single model only.<br>
 
| Range | Interpretation |
|---|---|
| < 0.1 | High confidence |
| 0.1 – 0.4 | Moderate uncertainty |
| 0.4 – 0.6 | Low confidence |
| > 0.6 | Very low confidence — model close to guessing |
| ≈ 0.693 | Maximum uncertainty — predicted probability ≈ 0.5 |
 
---
 
#### Adaptive Conformal Prediction Interval (regressors)<br>
 
Returned as a tuple `(lo, hi)` with 90% marginal coverage guarantee.<br>
 
We implement the **residual normalised conformity score** following [Lei et al. (2018)](https://doi.org/10.1080/01621459.2017.1307116) and [Cordier et al. (2023) / MAPIE](https://proceedings.mlr.press/v204/cordier23a.html). An auxiliary XGBoost model $\hat{\sigma}(\mathbf{x})$ is trained on held-out embeddings and absolute residuals |yᵢ − ŷᵢ|. At inference:<br>
 
$$[\hat{y}(\mathbf{x}) - q \cdot \hat{\sigma}(\mathbf{x}),\ \hat{y}(\mathbf{x}) + q \cdot \hat{\sigma}(\mathbf{x})]$$

 
where q is the ⌈(n+1)(1−α)⌉ / n quantile of the normalized scores sᵢ = |yᵢ − ŷᵢ| / σ̂(xᵢ).

 
- **Interval width varies per input** -- molecules more dissimilar to training data tend to receive wider intervals<br>

- **Coverage guarantee**: on exchangeable data, P(y ∈ [ŷ − qσ̂, ŷ + qσ̂]) ≥ 0.90<br>

- **The guarantee is marginal**, not conditional, as an unusually narrow interval on an out-of-distribution molecule does not guarantee correctness<br>

- **Full access**: We already computed MAPIE for all regression models; users are allowed to directly use them for customized model lists.<br>
 
---
 
#### Generating a MAPIE Bundle for a New Model<br>
 
To enable conformal uncertainty for a newly trained regression model:<br>
 
```bash
# Fit adaptive conformal bundle from val_predictions.csv
python fit_mapie_adaptive.py --root training_classifiers --prop <property_name>
```
 
The script reads `sequence`/`smiles` and `y_pred`/`y_true` columns from the CSV, recomputes embeddings, fits the XGBoost $\hat{\sigma}$ model, and saves `mapie_calibration.joblib` into the model directory. The bundle is automatically detected and loaded by `PeptiVersePredictor` on the next initialization.<br>
 


## Model Architecture 🌟

- **Sequence Embeddings:** [ESM-2 650M model](https://huggingface.co/facebook/esm2_t33_650M_UR50D) / [PeptideCLM model](https://huggingface.co/aaronfeller/PeptideCLM-23M-all) / [ChemBERTa](https://huggingface.co/DeepChem/ChemBERTa-77M-MLM). Foundational embeddings are frozen.
- **XGBoost Model:** Gradient boosting on pooled embedding features for efficient, high-performance prediction.
- **CNN/Transformer Model:** One-dimensional convolutional/self-attention transformer networks operating on unpooled embeddings to capture local sequence patterns.
- **Binding Model:** Transformer-based architecture with cross-attention between protein and peptide representations.
- **SVR Model:** Support Vector Regression applied to pooled embeddings, providing a kernel-based, nonparametric regression baseline that is robust on smaller or noisy datasets.
- **Others:** SVM and Elastic Nets were trained with [RAPIDS cuML](https://github.com/rapidsai/cuml), which requires a CUDA environment and is therefore not supported in the web app. Model checkpoints remain available in the Hugging Face repository.

## Troubleshooting 🌟

### LFS Download Issues

If files appear as SHA pointers:

```bash
huggingface-cli download ChatterjeeLab/PeptiVerse \
    training_data_cleaned/hemolysis/hemo_smiles_meta_with_split.csv \
    --local-dir . \
    --local-dir-use-symlinks False
```

## Citation 🌟

If you find this repository helpful for your publications, please consider citing our paper:

```
@article {Zhang2025.12.31.697180,
	author = {Zhang, Yinuo and Tang, Sophia and Chen, Tong and Mahood, Elizabeth and Vincoff, Sophia and Chatterjee, Pranam},
	title = {PeptiVerse: A Unified Platform for Therapeutic Peptide Property Prediction},
	elocation-id = {2025.12.31.697180},
	year = {2026},
	doi = {10.64898/2025.12.31.697180},
	publisher = {Cold Spring Harbor Laboratory},
	URL = {https://www.biorxiv.org/content/early/2026/01/03/2025.12.31.697180},
	eprint = {https://www.biorxiv.org/content/early/2026/01/03/2025.12.31.697180.full.pdf},
	journal = {bioRxiv}
}
```
To use this repository, you agree to abide by the MIT License.