Get started
Work

Pre-trained Deep Learning Models for Detecting Topological Threadings in Ring Polymers

Year: 2026
Type: software
Abstract: This repository contains pre-trained deep neural network models and an inference pipeline supporting the manuscript:"Learning to Detect Threadings in Melts of Rings" (Filippo Conforto, Roman Stano, Jan Smrek, and Davide Michieletto). INCLUDED MODELS:The archive contains four pre-trained neural network architectures in TensorFlow / Keras SavedModel format: 1. CNN1D_StA (models/NN_model_best_SIGWRITHE10_Norm_False_Net_CNN1D_Threading)- Input: 1D non-normalised Segment-to-All (StA) mutual writhe profile (shape: [batch, 200, 1]).- Performance: Primary model. Achieves >98% accuracy on test pairs and generalises without retraining to dense melts across chain lengths N = 100 to 1600 (TPR > 91%, FPR approx 0%). 2. RNN_StA (models/NN_model_best_SIGWRITHE10_Norm_False_Net_RNN_Threading)- Input: 1D non-normalised Segment-to-All (StA) writhe profile (shape: [batch, 200, 1]).- Role: Recurrent sequential classifier on 1D writhe representations. 3. CNN2D_StS (models/NN_model_best_SIGWRITHEMATRIX10_Norm_False_Net_CNN2D_Threading)- Input: 2D non-normalised Segment-to-Segment (StS) mutual writhe density matrix (shape: [batch, 100, 100, 1]).- Role: 2D convolutional network operating on the full pairwise mutual writhe field. 4. RNN_XYZ (models/NN_model_best_XYZ_Norm_True_Net_RNN_Threading)- Input: Normalised 3D Cartesian coordinates of the two ring polymers (shape: [batch, 200, 3]).- Role: Baseline coordinate-space recurrent model. CONTENTS:- models/: SavedModel directories for all 4 trained architectures.- load_and_predict.py: Self-contained Python script to load models and run verification/inference.- README.md: Quickstart instructions and input specifications.
DOI: 10.5281/zenodo.22893849
Language: English
Open Access status: green
OpenAlex ID: W7214034784
Cites:
Cited by:
Related to: 10