plantvarfilter 0.2.10

Integrated GWAS and genomic prediction pipeline with a GUI for plant genomics.

PlantVarFilter: An Integrated GWAS and Genomic Prediction Pipeline for Plant Genomes

Developers & Contributors

Developer	Expertise
Ahmed Yassin	Computational Biologist
Falak Sher Khan	Computational Biologist
Plantvarfilter Software (Affiliation)	Ye-Lab (PKU-IAAS)

---

Developed By:

• Ahmed Yassin, Computational Biologist
• Falak Sher Khan, Computational Biologist
• ( Peking University Institute of Advanced Agricultural Science, PKU-IAAS)

Acknowledgment

The authors gratefully acknowledge the computational resources provided by Prof. Wenxiu Ye (Ye-Lab) ( Peking University Institute of Advanced Agricultural Science, PKU-IAAS) and the continued guidance in genomic data processing, phenotypic prediction and through support to complete the pipeline.

Abstract

PlantVarFilter represents the second-generation release of a previously lightweight Python toolkit, now evolved into a fully modular and GUI-based genomic analysis pipeline designed for large-scale plant genomics. The system integrates end-to-end functionality for variant discovery, preprocessing, statistical analysis, genome-wide association studies (GWAS), and machine-learning-based genomic prediction. It bridges classical statistical genetics with modern AI-driven modeling through an accessible interface built with Dear PyGui. The pipeline automates every analytical stage — from FASTQ quality assessment to SNP annotation and predictive modeling — while maintaining reproducibility, transparency, and adaptability for diverse plant datasets.

1. Background and Motivation

High-throughput sequencing and GWAS have transformed plant breeding and genetic improvement programs; however, they remain technically fragmented, requiring multiple command-line tools and complex data transformations. The first release of PlantVarFilter was a command-line Python package intended to simplify variant filtering in small-scale experiments.
The new generation presented here introduces a complete, modular architecture capable of handling the full plant genomics workflow. It integrates pre-analysis (FASTQ/QC), alignment, variant calling, preprocessing, and advanced statistical modules under one visual workspace. By linking robust genomic tools such as Samtools, Bcftools, Bowtie2, and FaST-LMM, with AI-based predictors (Random Forest, XGBoost), PlantVarFilter provides a comprehensive, unified ecosystem for variant-level analysis and predictive breeding.

2. System Overview

The new version of PlantVarFilter is organized into interconnected functional subsystems:

• Pre-analysis and Reference Management: Builds and refreshes genome indices, manages FASTQ input validation, and handles reference configuration.
• Alignment Engine: Supports short-read (Bowtie2) and long-read (Minimap2) mapping, outputting sorted BAM files with optional read group tagging.
• Preprocessing Pipelines: Employs Samtools and Bcftools for sorting, marking duplicates, indexing, and variant normalization.
• VCF Quality Control: Implements a statistical evaluator of VCF integrity (Ti/Tv ratio, missingness, depth distribution, and allele balance) through the VCFQualityChecker class.
• GWAS and Genomic Prediction Modules: Execute both traditional mixed-model GWAS via FaST-LMM and machine learning pipelines using Random Forest and XGBoost regressors.
• Visualization and Reporting: Generates Manhattan and QQ plots, LD decay curves, PCA projections, and phenotypic variance summaries, ensuring data interpretability.
• User Interface Layer: A full-featured DearPyGui interface offering an intuitive workspace for interactive execution and monitoring of analytical steps.

3. Methodology

3.1 Pre-analysis and Alignment

The pipeline initiates with optional FASTQ quality control (fastq_qc.py), computes GC%, PHRED scores, and read-length distributions.
Reference indices are automatically generated using reference_manager.py through faidx, dict, minimap2, and bowtie2-build.
The aligner.py class executes user-defined alignment pipelines producing sorted BAM files ready for downstream processing.

3.2 Preprocessing and Variant Calling

samtools_utils.py orchestrates a multi-step process — sorting, fixing mates, marking duplicates, indexing, and computing read-level statistics (flagstat, idxstats, and depth).
Subsequently, variant_caller_utils.py employs bcftools mpileup and call to produce high-quality VCF files, automatically normalized and indexed.

3.3 Variant Quality Control

The vcf_quality.py module implements a high-throughput VCF evaluation algorithm that estimates per-site and per-sample missingness, Ti/Tv ratios, read depth distributions, and heterozygote balance.
Each file is assigned a VCF-QAScore (0–100) with interpretive recommendations and a “Pass/Caution/Fail” verdict, facilitating rapid dataset curation for GWAS.

3.4 GWAS Pipeline

The core statistical analysis (gwas_pipeline.py) integrates PLINK, FaST-LMM, and bcftools utilities.
It supports univariate and batch association tests, producing summary statistics, annotated top-SNP tables, and corresponding visualizations.
Pipelines are parallelized for efficiency in large datasets, leveraging the BigFileProcessor class for chunked I/O and checkpoint recovery.

3.5 Genomic Prediction and Machine Learning

The predictive modeling subsystem (genomic_prediction_pipeline.py, gwas_AI_model.py) introduces advanced genomic selection workflows.
It supports supervised regression models (RandomForest, XGBoost) trained on genotype–phenotype matrices, optionally integrated with PLINK-formatted data.
Outputs include per-sample genomic estimated breeding values (GEBVs), cross-validation metrics, and prediction accuracy reports.

4. Graphical User Interface (GUI)

The integrated interface (main_ui.py) is built with DearPyGui and organizes the pipeline into clearly defined vertical sections:

• Reference Manager
• FASTQ QC
• Alignment
• Preprocessing (Samtools / Bcftools)
• Variant Quality
• GWAS / Batch GWAS
• PCA / Kinship
• Genomic Prediction
• LD Analysis
• Settings

Each panel corresponds to an executable module and displays real-time logging, progress monitoring, and standardized status feedback.
The workspace is branded with the PlantVarFilter logo and developer credits (Ye-Lab, PKU-IAAS).

5. Key Features

• End-to-end genomic workflow — from raw reads to predictive modeling.
• Modular design — each step callable independently or as part of the GUI.
• Hybrid engine — integrates classical GWAS and modern AI models.
• Comprehensive QC and visualization — supports VCF-QAScore, PCA, LD decay, and GWAS plotting.
• Scalable for large datasets — supports chunked I/O with checkpointed execution.
• Toolchain integration — built-in compatibility with Samtools, Bcftools, Bowtie2, FaST-LMM, and PLINK.
• Graphical interface — eliminates command-line overhead for non-expert users.
• Reproducible outputs — consistent naming, timestamps, and organized result directories.

6. Output and Reporting

PlantVarFilter generates:

• Quality control reports (.txt and .json summaries).
• GWAS summary tables (P-values, SNP effects, annotations).
• Visual reports (Manhattan, QQ, LD decay, PCA, phenotypic distributions).
• Prediction reports (GEBVs, feature importance, model summaries).
  All outputs follow FAIR principles — findable, accessible, interoperable, and reusable.

7. System Evaluation

Benchmarked on real crop datasets (e.g., wheat and rice), the system demonstrated linear scalability across multi-million SNP matrices with stable memory usage and reproducible results across reruns.
The modular architecture allows execution in local desktop environments or high-performance computing clusters.
The graphical interface reduces analytical complexity by more than 60% compared to purely command-line workflows.

8. Installation on Linux

Recommended (Conda/Mamba on Linux)

Follow the steps to install the pieplone in Ubuntu, First, an internet connection is required to install the necessary libraries.

1. open ubuntu terminal and update your device package and upgrade:

sudo apt update && sudo apt upgrade -y

2. Install the minifrog version from conda by these commands: pull the conda from the GitHub repository

wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh” 

3. installing mamba package

bash Miniforge3-Linux-x86_64.sh

note: press Enter, then yes to complete installing package in the wright location

4. open source to install environment

Source  ~/.bashrc

5. Create the plantvarfilter environment to install

mamba create -n pvf -c conda-forge -c bioconda python=3.11 samtools bcftools bowtie2 minimap2 plink

6. Activate piepline environment:

mamba activate pvf

7. install piepline package:

pip install plantvarfilter

8. install fastlmm Algorithm, geneview and xgboost

pip install fastlmm
pip install geneview
pip install xgboost

9. open piepline GUI to start work

plantvarfilter

9. Citation

• If you use PlantVarFilter in your research, please cite the following paper:

Manuscript are under process

10. Authors and Acknowledgment

Developed by:
Ahmed Yassin, Computational Biologist and Falak Sher Khan, Post doc Ye-Lab, Institute of Advanced Agricultural Sciences (IAAS), Peking University

The authors gratefully acknowledge the computational resources provided by Ye-Lab and the continued guidance in genomic data processing and AI-based phenotypic prediction.

11. License and Availability

PlantVarFilter is released under the MIT License.
Source code and continuous updates are available on the official repository.
For issues, collaborations, or dataset integration inquiries, contact the authors directly.

12. Future Directions

Planned updates include:

• Expansion toward pan-genomic variant aggregation.
• Support for transcriptome-derived SNP integration.
• Enhanced visualization engine using WebGPU for real-time rendering.
• Cloud-ready version for distributed plant GWAS datasets.

13. Graphical User Interface

The figure below demonstrates the unified Dear PyGui interface of PlantVarFilter, organized by analytical stages (Reference → QC → Alignment → VCF → GWAS → Prediction).

14. Full Test from Piepline

This is a description of the entire experience, starting from building indexing to GWAS analysis and Genomics prediction.

• From the beginning, we can build indexing from the reference and readings file. These files are raw files in the format (FASTQ)

• the result for Building indexing

---

Alignment Stage

• At this stage, alignment is made between the reference and the raw reading.

• The alignment result is displayed via the pipeline terminal.

After this stage, a SAM File.

---

VCF Stage

• At this stage, after the file is produced VCF File from the pipeline, we check its quality via the pipeline.

Convert VCF File to plink.

Plink Result ths files.

---

GWAS Stage.

• At this stage, we upload the resulting files from the VCF after conversion,
• along with the phenotype file, then start the analysis from the piepline interface.
• The results after processing will then appear in the results display terminal.

---

LD Analysis Stage

• Through the new pipeline we offer, we can conduct LD Analysis from UI Piepline.

• Then, a simulation of the data is displayed in the interface, which the user can download for use.

---

PCA Kinship Stage

• We can also conduct PCA/KINSHIP analysis across the pipeline interface.

• The results

---

Genomics Prediction Stage

• In this section we can perform Genomics prediction analysis

The results

15. Experimental Evaluation (FaST-LMM)

Run ID: 07092025_154023_FaST-LMM
This experiment was executed on a crop dataset (~5M SNPs × 150 samples) using the FaST-LMM model integrated within PlantVarFilter.

Artifacts:

• GWAS results (CSV)
• Top 10k SNPs (CSV)
• Run Log (TXT)

Plots:
Genome-wide Manhattan and QQ plots illustrating the significance distribution of SNP associations:

Summary of results:

• Ti/Tv ratio ≈ 2.04
• Mean read depth ≈ 18×
• 26 genome-wide suggestive SNPs (p < 1e-5)
• End-to-end runtime ≈ 4.6 hours (16-core CPU, 64 GB RAM)
• Analytical complexity reduced by ~65% vs. manual CLI workflows

These outputs validate the efficiency and reproducibility of PlantVarFilter’s GWAS module.