faithlola/demographic-Inference

Machine Learning Approaches for Demographic Inference in African Malaria Vector Populations # demographic-Inference # MSc Bioinformatics Thesis: Machine Learning Approaches for Demographic Inference in African Malaria Vector Populations This repository contains code, scripts, and results from my MSc Bioinformatics thesis on **evolutionary dynamics and genomic surveillance of mosquito populations**, with a focus on **site frequency spectrum (SFS) analysis, msprime simulations, and machine learning approaches** for inferring effective population size (Ne). --- ## 📂 Repository Structure - **`sfs.py`** Core functions for computing **site frequency spectra (SFS)** from simulated or VCF-based data. - Generates folded/unfolded spectra. - Normalises vectors for input into machine learning models. - **`msprime.py`** Scripts for **simulating genomic data** under neutral demographic models using `msprime`. - Simulates allele frequency data under broad and narrow Ne schemes. - Produces SFS vectors and plots for downstream ML analysis. - **`svm.py`** Implements a **Support Vector Machine (SVM) classifier** for Ne inference. - Input: Folded, normalised SFS vectors. - Output: Predicted Ne class (broad or narrow). - Uses RBF kernel, trained with 80/20 train-test split. - **`random-forest-regressor.py`** Implements a **Random Forest regressor** for continuous Ne estimation. - Input: Folded, normalised SFS vectors. - Output: Continuous Ne predictions. - Includes a mapping step to nearest class bin for direct accuracy comparison with classifiers. - **`Permutation-Scenarios.py`** Performs **paired permutation testing** to compare SVM and Random Forest classifiers. - Runs both broad (100k, 500k, 1M) and narrow (50k, 225k, 500k) Ne schemes. - Produces permutation null distributions, p-values, and confusion matrices. - **`README.md`** Documentation and usage guide for the repository. --- ## 🧬 Data Preparation Population-specific mosquito sample IDs were extracted from the *Anopheles gambiae 1000 Genomes Project (Ag1000G, Phase 2 chromosome 3L variant set)* metadata f …