Reconstruction of Masked DNA Sequences

This repository contains the prediction phase of the research paper: "Reconstruction of Masked Sequences via Inverse Mapping of Incomplete Information Natural Vectors".

Overview

This project tackles a key challenge in bioinformatics: reconstructing complete biological sequences from their numerical representations, especially when parts of the original sequence are missing or "masked."

Our approach involves two main steps:

1. Feature Extraction: We introduce a novel method called Incomplete Information Natural Vectors to convert masked DNA sequences into fixed-dimensional numerical vectors while preserving crucial positional information.
2. Sequence Reconstruction: We utilize a pre-trained Long Short-Term Memory (LSTM) neural network that performs the inverse mapping, successfully reconstructing the original 16-nucleotide sequence from its natural vector with high accuracy.

The model was trained and validated on genomic subsequence datasets from SARS-CoV-2 and HIV-1.

Main.py

This notebook demonstrates the prediction capabilities of the trained LSTM model. It performs the following steps:

1. Loads Data: Reads a CSV file containing pre-calculated incomplete information natural vectors for a set of masked DNA subsequences.
2. Loads Model: Loads the pre-trained LSTM model from the saved_model/ directory.
3. Predicts Sequences: Uses the model to predict the original 16-nucleotide sequences from their vector representations.
4. Evaluates Accuracy: Compares the reconstructed sequences to the original sequences to calculate the prediction accuracy.
5. Saves Results: Outputs the predictions and accuracy metrics to a new CSV file.

How to Run

Prerequisites

• Python 3.10.15
• Required libraries: pandas, numpy, tensorflow

You can install the dependencies using pip:

pip install pandas numpy tensorflow