OSCR

AUKAT: Conditional VAE-Driven Augmentation and Neural Modeling of Enzyme Turnover Numbers.

Code ↔ Paper

7 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 7 matches
  1. [1] § 2. Materials and Methods › 2.8. Implementation Details and Reproducibility ↔ train.py, lines 120–236 · score 0.86 · AdamW, weight decay, squared error, neural network, NumPy, epochs
  2. [2] § 2. Materials and Methods › 2.7. Evaluation Metrics ↔ predict_kcat.py, lines 155–175 · score 0.62 · squared error, ground truth, power, sum, linear, metric
  3. [3] § 2. Materials and Methods › 2.4. Neural Network Architecture for Prediction ↔ predict_kcat.py, lines 37–90 · score 0.55 · Conv1d, flattened, ReLU, module, linear, layers
  4. [4] § 2. Materials and Methods › 2.4. Neural Network Architecture for Prediction ↔ predict_kcat.py, lines 37–90 · score 0.54 · transformer encoder layers, flattened, module, Network, Neural, embeddings
  5. [5] § 2. Materials and Methods › 2.1. Datasets and Preprocessing › 2.1.1. Curated Dataset ↔ train.py, lines 30–67 · score 0.54 · ec2vec, mol2vec, blocks, concatenated, species, embeddings
  6. [6] § 2. Materials and Methods › 2.6. Feature Importance Analysis ↔ train.py, lines 30–67 · score 0.50 · ec2vec, mol2vec, block, species, embeddings, training
  7. [7] § 2. Materials and Methods › 2.4. Neural Network Architecture for Prediction ↔ model.py, lines 7–51 · score 0.50 · Conv1d, ReLU, module, linear, layers, network

Paper

Loaded from Europe PMC by your browser, not stored by OSCR: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The authors' code

Python · 240 lines · 7.4 KB · no license · 3 matches

  1. # train.py
  2. import argparse
  3. import os
  4. from typing import List, Tuple
  6. import numpy as np
  7. import pandas as pd
  8. import torch
  9. import torch.nn as nn
  10. import torch.optim as optim
  11. from scipy.stats import pearsonr
  12. from sklearn.model_selection import train_test_split
  13. from sklearn.metrics import mean_squared_error, r2_score
  14. from torch.utils.data import DataLoader, TensorDataset
  16. from model_for_github import NeuralNetwork
  19. def natural_sort_key(s: str):
  20. """
  21. Sort keys like mol2vec_0, mol2vec_1, ..., mol2vec_10 numerically.
  22. Falls back to string sort if no trailing integer is found.
  23. """
  24. try:
  25. return (0, int(s.split("_")[-1]))
  26. except Exception:
  27. return (1, s)
  30. def detect_feature_columns(df: pd.DataFrame) -> List[str]:
  31. """
  32. Auto-detect feature columns in the canonical order:
  33. mol2vec_* then ec2vec_* then Embedding_*
  35. This matches your original concatenation order (metabolite-EC-species).
  36. """
  37. mol_cols = [c for c in df.columns if c.startswith("mol2vec_")]
  38. ec_cols = [c for c in df.columns if c.startswith("ec2vec_")]
  39. emb_cols = [c for c in df.columns if c.startswith("Embedding_")]
  41. mol_cols = sorted(mol_cols, key=natural_sort_key)
  42. ec_cols = sorted(ec_cols, key=natural_sort_key)
  43. emb_cols = sorted(emb_cols, key=natural_sort_key)
  45. feature_columns = mol_cols + ec_cols + emb_cols
  47. if len(feature_columns) == 0:
  48. raise ValueError(
  49. "No feature columns found. Expected columns starting with "
  50. "'mol2vec_', 'ec2vec_', and/or 'Embedding_'."
  51. )
  53. print("Detected feature columns:")
  54. print(f" mol2vec_* : {len(mol_cols)}")
  55. print(f" ec2vec_* : {len(ec_cols)}")
  56. print(f" Embedding_* : {len(emb_cols)}")
  57. print(f" TOTAL : {len(feature_columns)}")
  59. # Helpful warning if one block is missing (still allowed)
  60. if len(mol_cols) == 0:
  61. print("WARNING: No mol2vec_* columns detected.")
  62. if len(ec_cols) == 0:
  63. print("WARNING: No ec2vec_* columns detected.")
  64. if len(emb_cols) == 0:
  65. print("WARNING: No Embedding_* columns detected.")
  67. return feature_columns
  70. def make_loaders(
  71. X_train: np.ndarray,
  72. y_train: np.ndarray,
  73. X_val: np.ndarray,
  74. y_val: np.ndarray,
  75. batch_size: int,
  76. ) -> Tuple[DataLoader, DataLoader]:
  77. X_train_t = torch.tensor(X_train, dtype=torch.float32)
  78. y_train_t = torch.tensor(y_train, dtype=torch.float32)
  79. X_val_t = torch.tensor(X_val, dtype=torch.float32)
  80. y_val_t = torch.tensor(y_val, dtype=torch.float32)
  82. train_loader = DataLoader(
  83. TensorDataset(X_train_t, y_train_t),
  84. batch_size=batch_size,
  85. shuffle=True,
  86. drop_last=False,
  87. )
  88. val_loader = DataLoader(
  89. TensorDataset(X_val_t, y_val_t),
  90. batch_size=batch_size,
  91. shuffle=False,
  92. drop_last=False,
  93. )
  94. return train_loader, val_loader
  97. def parse_args():
  98. p = argparse.ArgumentParser(
  99. description="Train CNN+Transformer kcat predictor (single train/val split, no pseudo data)."
  100. )
  102. p.add_argument("--data_csv", type=str, required=True, help="CSV path.")
  103. p.add_argument("--target_column", type=str, default="kcat")
  105. p.add_argument("--save_dir", type=str, default="saved_NN_models")
  106. p.add_argument("--ckpt_name", type=str, default="best_model.pth")
  108. p.add_argument("--val_ratio", type=float, default=0.2)
  109. p.add_argument("--seed", type=int, default=42)
  111. p.add_argument("--epochs", type=int, default=300)
  112. p.add_argument("--batch_size", type=int, default=32)
  113. p.add_argument("--lr", type=float, default=2e-4)
  114. p.add_argument("--weight_decay", type=float, default=5e-3)
  116. p.add_argument("--no_cuda", action="store_true")
  117. return p.parse_args()
  120. def main():
  121. args = parse_args()
  123. device = torch.device("cpu" if args.no_cuda or not torch.cuda.is_available() else "cuda")
  124. print(f"Using device: {device}")
  126. # reproducibility
  127. torch.manual_seed(args.seed)
  128. np.random.seed(args.seed)
  130. df = pd.read_csv(args.data_csv)
  132. # auto-detect features in metabolite-EC-species order
  133. feature_columns = detect_feature_columns(df)
  135. # required target
  136. if args.target_column not in df.columns:
  137. raise KeyError(f"Target column '{args.target_column}' not found in CSV.")
  139. # drop NaNs in features + target
  140. df = df.dropna(subset=feature_columns + [args.target_column]).reset_index(drop=True)
  142. # split
  143. train_df, val_df = train_test_split(
  144. df, test_size=args.val_ratio, random_state=args.seed, shuffle=True
  145. )
  146. print(f"Split: train={len(train_df)} val={len(val_df)}")
  148. # numpy arrays
  149. X_train = train_df[feature_columns].values
  150. y_train = train_df[args.target_column].astype(np.float64).values
  152. X_val = val_df[feature_columns].values
  153. y_val = val_df[args.target_column].astype(np.float64).values
  155. # log10 transform (consistent with your original code)
  156. # NOTE: kcat must be > 0
  157. if np.any(y_train <= 0) or np.any(y_val <= 0):
  158. raise ValueError("Found non-positive kcat values. log10 requires kcat > 0.")
  160. y_train_log = np.log10(y_train)
  161. y_val_log = np.log10(y_val)
  163. train_loader, val_loader = make_loaders(X_train, y_train_log, X_val, y_val_log, args.batch_size)
  165. # model
  166. input_size = X_train.shape[1]
  167. model = NeuralNetwork(input_size).to(device)
  168. optimizer = optim.AdamW(model.parameters(), lr=args.lr, weight_decay=args.weight_decay)
  169. criterion = nn.MSELoss()
  171. os.makedirs(args.save_dir, exist_ok=True)
  172. ckpt_path = os.path.join(args.save_dir, args.ckpt_name)
  174. best_val_loss = float("inf")
  175. best_metrics = None
  177. for epoch in range(args.epochs):
  178. # train
  179. model.train()
  180. train_loss_sum = 0.0
  182. for xb, yb in train_loader:
  183. xb = xb.to(device)
  184. yb = yb.to(device)
  186. optimizer.zero_grad(set_to_none=True)
  187. pred = model(xb).squeeze()
  188. loss = criterion(pred, yb)
  189. loss.backward()
  190. optimizer.step()
  192. train_loss_sum += loss.detach().cpu().item()
  194. avg_train_loss = train_loss_sum / max(len(train_loader), 1)
  196. # val
  197. model.eval()
  198. val_loss_sum = 0.0
  199. preds = []
  201. with torch.no_grad():
  202. for xb, yb in val_loader:
  203. xb = xb.to(device)
  204. yb = yb.to(device)
  206. pred = model(xb).squeeze()
  207. loss = criterion(pred, yb)
  208. val_loss_sum += loss.detach().cpu().item()
  209. preds.extend(pred.detach().cpu().numpy())
  211. avg_val_loss = val_loss_sum / max(len(val_loader), 1)
  213. y_pred = np.array(preds, dtype=np.float64)
  214. mse = mean_squared_error(y_val_log, y_pred)
  215. r2 = r2_score(y_val_log, y_pred)
  216. pear = pearsonr(y_val_log, y_pred)[0]
  218. if (epoch + 1) % 10 == 0 or epoch == 0:
  219. print(
  220. f"epoch {epoch+1:4d}/{args.epochs} "
  221. f"train_loss={avg_train_loss:.6f} val_loss={avg_val_loss:.6f} "
  222. f"MSE={mse:.6f} R2={r2:.6f} Pearson={pear:.6f}"
  223. )
  225. if avg_val_loss < best_val_loss:
  226. best_val_loss = avg_val_loss
  227. best_metrics = (mse, r2, pear)
  228. torch.save(model.state_dict(), ckpt_path)
  230. print("\n=== Best checkpoint (by lowest val loss) ===")
  231. print(f"Saved to: {ckpt_path}")
  232. if best_metrics is not None:
  233. mse, r2, pear = best_metrics
  234. print(f"Val MSE: {mse:.6f}")
  235. print(f"Val R2: {r2:.6f}")
  236. print(f"Val Pearson: {pear:.6f}")
  239. if __name__ == "__main__":
  240. main()

train.py at commit 78a2a6c, no license · at the source

Overview

Authors: Mengmeng Liu1, Xialong Ni2, Michal Brylinski1,2
ORCID iDs: Mengmeng Liu, Xialong Ni, Michal Brylinski
  1. Center for Computation and Technology, Louisiana State University, Baton Rouge, LA 70803, USA
  2. Department of Biological Sciences, Louisiana State University, Baton Rouge, LA 70803, USA
Institutions: Louisiana State University (United States)
Journal: Biomolecules, volume 16, issue 7, article 1049
Dates: received 27 May 2026; accepted 16 July 2026; published online 18 July 2026
Type: Research article · Language: English
License: CC BY
Identifiers: DOI 10.3390/biom16071049 · PMID 42509841 · PMCID PMC13406586 · OpenAlex W7169723205
Open access: gold, a free copy (OpenAlex)
Status: code verified
Categories: human (organism)
Methods: Statistics, Machine learning, Connectivity
Keywords: kcat prediction, enzyme kinetics, conditional variational autoencoder, synthetic data augmentation, deep learning, transformer models
MeSH: Enzymes*, Neural Networks, Computer*, Autoencoder, Generative Artificial Intelligence, Kinetics (* major topic)
Topic: Bioinformatics and Genomic Networks (Molecular Biology, Biochemistry, Genetics and Molecular Biology), according to OpenAlex
Funding: Center for Computation and Technology at Louisiana State Universit
Citations: not cited yet (Europe PMC); 34 references in the paper

Abstract

Accurate prediction of enzyme turnover numbers (kcat) is essential for applications in systems biology, metabolic engineering, and drug discovery, yet remains challenging due to the limited availability and uneven distribution of experimental data. Here, we present AUKAT, an integrated framework that combines conditional generative modeling with deep neural prediction to improve kcat estimation. A conditional variational autoencoder generates synthetic training instances in embedding space, followed by a selection pipeline that retains samples with strong agreement across independent evaluators, thereby ensuring data reliability. A hybrid convolutional neural network and transformer-based architecture is then used to predict kcat from substrate, enzyme functional, and species embeddings. Incorporating synthetic data improved predictive performance for both random forest and neural network models in five-fold cross-validation, with larger gains observed for the neural network architecture. Benchmarking against DLKcat demonstrated comparable predictive accuracy on the standard test set, while evaluation on stricter unseen subsets indicated improved generalization for low-similarity substrates and enzymes. Feature importance analysis further showed that AUKAT leverages substrate, enzyme functional, and species information in a more balanced manner rather than relying predominantly on a single feature source. In addition, AUKAT-human, a specialized model trained using a pre-training and fine-tuning strategy, achieved improved prediction accuracy for human enzyme kinetics. Overall, AUKAT provides a scalable approach for enzyme kinetics prediction and offers a practical solution to data scarcity in biochemical modeling.

Reproduced under the paper's license (CC BY), from the paper cited above.

Repository

Its files are read in the Code ↔ Paper reader above, with 7 matches between paragraphs and lines of code.

MengLiu90/AUKCAT-Neural-Network-Model-for-Kcat-Prediction

License: none: the authors keep all their rights
State: the link answers, verified on 27 September 2026
Evidence: files inventoried
Commit: 78a2a6c14cfb4a4b805544d20cfec0fb8a7072ce, 24 June 2026
Languages: Python (3)
Size: 27 files, 3 scripts
Software Heritage: not archived
Found in: “Data Availability Statement”
Holds: README
Not found: license file, CITATION.cff, environment file, tests, continuous integration, documentation
Tools: PyTorch (3 files), NumPy (2 files), pandas (2 files), scikit-learn (2 files), SciPy (2 files)
Availability: 1 check, the latest on 27 September 2026: the link answers
  • 27 September 2026: the link answers
4 files
At the source: github.com/MengLiu90/AUKCAT-Neural-Network-Model-for-Kcat-Prediction

The paper's code and data availability statement is in the Data section.

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

  • 1 repository of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
  • 3 scripts, each with its path and the digest of its content;
  • 7 matches between paragraphs of the paper and lines of the code (method lexical-v1);
  • neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

No dataset and no data link were found in the paper.

Data Availability Statement

The code and data are available at https://github.com/MengLiu90/AUKCAT-Neural-Network-Model-for-Kcat-Prediction, (accessed on 24 June 2026).

Reproduced under the paper's license (CC BY), from the paper cited above.

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 1, 27 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 3 authors, 6 keywords, 5 MeSH terms, 1 funder, 29 references.

Cite

This paper

Liu, M., Ni, X., & Brylinski, M. (2026). AUKAT: Conditional VAE-Driven Augmentation and Neural Modeling of Enzyme Turnover Numbers. Biomolecules, 16(7), 1049. https://doi.org/10.3390/biom16071049

BibTeX

@article{liu2026aukat,
author = {Liu, Mengmeng and Ni, Xialong and Brylinski, Michal},
title = {{AUKAT: Conditional VAE-Driven Augmentation and Neural Modeling of Enzyme Turnover Numbers}},
journal = {Biomolecules},
year = {2026},
month = jul,
volume = {16},
number = {7},
pages = {1049},
publisher = {Multidisciplinary Digital Publishing Institute (MDPI)},
issn = {2218-273X},
doi = {10.3390/biom16071049},
url = {https://doi.org/10.3390/biom16071049},
pmid = {42509841},
pmcid = {PMC13406586}
}

RIS

TY - JOUR
AU - Liu, Mengmeng
AU - Ni, Xialong
AU - Brylinski, Michal
TI - AUKAT: Conditional VAE-Driven Augmentation and Neural Modeling of Enzyme Turnover Numbers
T2 - Biomolecules
J2 - Biomolecules
PY - 2026
DA - 2026/07/18
VL - 16
IS - 7
SP - 1049
SN - 2218-273X
PB - Multidisciplinary Digital Publishing Institute (MDPI)
DO - 10.3390/biom16071049
UR - https://doi.org/10.3390/biom16071049
LA - en
ER -

CSL-JSON

{
"id": "10.3390/biom16071049",
"type": "article-journal",
"title": "AUKAT: Conditional VAE-Driven Augmentation and Neural Modeling of Enzyme Turnover Numbers",
"container-title": "Biomolecules",
"author": [
{
"family": "Liu",
"given": "Mengmeng"
},
{
"family": "Ni",
"given": "Xialong"
},
{
"family": "Brylinski",
"given": "Michal"
}
],
"container-title-short": "Biomolecules",
"volume": "16",
"issue": "7",
"page": "1049",
"DOI": "10.3390/biom16071049",
"PMID": "42509841",
"PMCID": "PMC13406586",
"ISSN": "2218-273X",
"publisher": "Multidisciplinary Digital Publishing Institute (MDPI)",
"URL": "https://doi.org/10.3390/biom16071049",
"language": "en",
"issued": {
"date-parts": [
[
2026,
7,
18
]
]
}
}

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Similar papers

The papers with a page that share the most with this one: the tools found in their code, their categories, datasets, cited references and authors, the rarest counting most.

[1] doi:10.1002/pro.70695 [code]
MIF-MAPMS: Enhancing identification of myelin autoantigenic peptides in multiple sclerosis through multimodal information fusion.
Journal: Protein science : a publication of the Protein Society
In common: PyTorch, scikit-learn, pandas, 1 other tool, 2 references
[2] doi:10.1002/advs.202523984 [code]
INB&lt;sup&gt;3&lt;/sup&gt;P: A Multi-Modal and Interpretable Co-Attention Framework Integrating Property-Aware Explanations and Memory-Bank Contrastive Fusion for Blood-Brain Barrier Penetrating Peptide Discovery.
Journal: Advanced science (Weinheim, Baden-Wurttemberg, Germany)
In common: PyTorch, scikit-learn, pandas, 2 other tools, 1 reference
[3] doi:10.1038/s41598-026-61605-4 [code]
Learning precise segmentation of neurofibrillary tangles from rapid manual point annotations.
Journal: Scientific reports
In common: PyTorch, scikit-learn, pandas, 1 other tool, 1 reference
[4] doi:10.3390/brainsci16050455 [code]
Spec-RWKV: A Spectrum-Guided Multi-Scale Recurrent Modeling Framework for Multi-Center Resting-State fMRI-Assisted Diagnosis.
Journal: Brain sciences
In common: PyTorch, scikit-learn, pandas, 1 other tool, 1 reference
[5] doi:10.1186/s13321-026-01191-9 [code]
Multiscale analysis and optimal glioma therapeutic candidate discovery using the CANDO platform.
Journal: Journal of cheminformatics
In common: scikit-learn, pandas, SciPy, 1 other tool, 1 reference
[6] doi:10.1021/acs.jmedchem.5c03413 [code]
Peptides Targeting GDNF Family Receptor Alpha 1 (GFRα1) Mimic Glial Cell Line-Derived Neurotrophic Factor (GDNF) Bioactivity.
Journal: Journal of medicinal chemistry
In common: scikit-learn, pandas, SciPy, 1 other tool, 1 reference
[7] doi:10.1038/s41467-026-71406-y [code]
CellLoop: Identifying single-cell 3D genome chromatin loops.
Journal: Nature communications
In common: scikit-learn, pandas, SciPy, 1 other tool, 1 reference
[8] doi:10.1002/dad2.70406 [code]
Deep contrastive learning framework identifies cell-type-specific drug targets in Alzheimer's disease.
Journal: Alzheimer's & dementia (Amsterdam, Netherlands)
In common: PyTorch, scikit-learn, NumPy, 1 reference
[9] doi:10.1038/s41467-026-75806-y [code]
Cryo-EM insights into isoform-specific properties of the IP&lt;sub&gt;3&lt;/sub&gt;R2 channel.
Journal: Nature communications
In common: pandas, SciPy, NumPy, 1 reference
[10] doi:10.3389/fbinf.2026.1858866 [code]
Comparative evaluation of reference-free transcriptomic deconvolution highlights the importance of biological validation in astrocytes across Alzheimer's disease.
Journal: Frontiers in bioinformatics
In common: 1 reference

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: questions and error reports about this paper and its code, from signed-in readers and its authors. It opens with sign-in.

Reproductions: reports from readers who ran the authors' code: what they reproduced, with which environment, commit and data. It opens with sign-in.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.