Bioinformatics Tools and Genomics Pipelines

Introduction to Bioinformatics

Bioinformatics is the discipline applying computational methods to store, retrieve, analyse, and interpret large-scale biological data—particularly sequence data, protein structures, and genomic information. The discipline emerged from the need to handle the data generated by DNA sequencing: the first GenBank submission (1982) contained 606 bases; by 2024 GenBank contains over 2.5 trillion base pairs. Modern whole-genome sequencing generates 30× coverage of the 3.2 gigabase human genome in hours, producing hundreds of gigabytes per sample that require sophisticated computational pipelines for quality control, read alignment, variant calling, and functional annotation before clinical or research interpretation.

Key databases structuring biological knowledge include NCBI (GenBank sequences, PubMed literature, dbSNP variants, ClinVar clinical variants), Ensembl (annotated genome assemblies), UniProt (protein sequences and function), PDB (Protein Data Bank—protein and nucleic acid 3D structures), KEGG (metabolic pathways), and Gene Ontology (standardised gene function annotation). The FAIR principles (Findable, Accessible, Interoperable, Reusable) guide modern biological data management, enabling reproducible research and meta-analysis across studies and species.

Sequence Alignment and Genome Assembly

Short Read Alignment

Illumina sequencing produces millions of short reads (75-300 bp) that must be aligned to a reference genome. BWA-MEM2 and Bowtie2 use Burrows-Wheeler transform and FM-index data structures enabling near-exact alignment of 50+ million reads to the 3.2 Gb human genome within minutes. Alignment produces SAM/BAM files containing read positions, quality scores, and alignment flags. STAR and HISAT2 are splice-aware aligners for RNA-seq—accounting for reads spanning exon-exon junctions. Post-alignment processing: samtools for sorting/indexing/filtering, Picard MarkDuplicates reducing PCR duplicate bias, GATK BaseRecalibration adjusting systematic error rates by read position and nucleotide context—each step essential for downstream variant calling accuracy. CIGAR strings in BAM files encode alignment shape: matches, deletions, insertions, soft clips.

Long Read Assembly

PacBio (HiFi) and Oxford Nanopore long reads (1-100+ kb) enable spanning repetitive regions, resolving structural variants, and phasing haplotypes that short reads cannot. De novo genome assembly from long reads uses overlap-layout-consensus (OLC) assemblers: Hifiasm (PacBio HiFi), Flye (Nanopore)—constructing assembly graphs by finding read overlaps and resolving them into contiguous sequences. T2T-CHM13 (telomere-to-telomere) human genome assembly (2022) added 200+ Mb of previously unresolved sequence including all centromeres, telomeres, and segmental duplications—completing the human genome reference 20 years after the original draft. Hybrid assembly combining short and long reads (short read polishing of long read backbone) achieves highest accuracy.

Variant Calling and Annotation

SNV and Indel Calling

GATK HaplotypeCaller is the standard tool for germline SNV and small indel detection: local de novo assembly in active regions from realigned reads, followed by statistical genotype likelihood calculation using position-specific models, producing VCF (variant call format) files with genotype calls and quality metrics (GQ, DP, FILTER). VQSR (variant quality score recalibration) uses a Gaussian mixture model trained on known variant databases to classify variants as PASS or filtered. Somatic variant calling (tumour-normal paired samples): GATK Mutect2 identifies tumour-specific variants using matched normal as panel-of-normals background; CNVkit and FACETS identify copy number variants. Standard accuracy benchmarking on NIST/Genome in a Bottle truth sets enables comparison between callers and validates clinical pipeline performance.

RNA-seq Analysis

RNA-seq workflow: raw reads quality checked with FastQC; adaptor-trimmed with Trimmomatic or Cutadapt; aligned with STAR or pseudo-aligned with Salmon/kallisto (quantifying transcript abundances directly from reads without alignment); count matrices generated with featureCounts or HTSeq-count; differential expression with DESeq2 or edgeR (negative binomial models for overdispersed count data with sample size-appropriate normalisation); pathway enrichment with fgsea or clusterProfiler. Alignment-free quantification (Salmon/kallisto) reduces computational time 100-fold with equivalent accuracy. Multi-sample integration (batch correction with ComBat or limma removeBatchEffect) is essential for cross-study meta-analyses. ENCODE and GTEx consortia provide reference compendiums of gene expression across tissues and conditions.

Structural Bioinformatics

AlphaFold2 (DeepMind, 2021) predicts protein 3D structure from primary sequence with near-experimental accuracy using co-evolutionary multiple sequence alignment features and attention-based deep learning—solving a 50-year grand challenge in computational biology. The AlphaFold protein structure database now contains predictions for essentially all ~200 million known proteins. AlphaFold-Multimer extends predictions to protein complexes; AlphaFold3 (2024) predicts structures of proteins with DNA, RNA, and small molecule ligands. Rosetta ab initio structure prediction, molecular dynamics (GROMACS, AMBER, NAMD), and molecular docking (AutoDock, Glide) complement experimental structures in drug discovery for identifying binding pockets, optimising lead compounds, and understanding protein-drug interactions.

Examples and Applications

Example 1: Clinical Whole Exome Sequencing Pipeline

Clinical WES for rare disease diagnosis: DNA QC → library preparation → exome capture (Agilent SureSelect, IDT xGen) → Illumina sequencing → BWA-MEM alignment → GATK HaplotypeCaller → VEP/ANNOVAR functional annotation → filtering by MAF (minor allele frequency below 0.01 in gnomAD), predicted pathogenicity (CADD score >20), disease gene lists (OMIM, ClinVar pathogenic) → trio analysis (parents sequenced simultaneously enables de novo variant identification) → report generation for clinical interpretation. Average diagnostic yield 25-35% for undiagnosed genetic disease, rising to 40-50% when phenotyping is precise. Cases unresolved after WES can proceed to whole genome sequencing detecting non-coding variants, structural variants, and repeat expansions missed in exome.

Example 2: GWAS Bioinformatics

GWAS (genome-wide association studies) test millions of SNPs for association with phenotype across hundreds of thousands of individuals. Quality control pipeline: genotyping array data → sample QC (call rate, sex check, outlier removal) → SNP QC (MAF filter, HWE departure, call rate) → ancestry principal component analysis (PCA) → imputation (using Haplotype reference consortium panel expanding 700k array SNPs to ~10M imputed variants) → association test (PLINK2 logistic/linear regression including PCs as covariates) → meta-analysis of multiple cohorts (METAL) → LD-clumping to identify independent association signals → fine-mapping (SuSiE, FINEMAP polyfine) to prioritise causal variants → colocalization with eQTLs identifying likely causal genes.

Example 3: Single-Cell RNA-seq Pipeline

scRNA-seq bioinformatics: raw reads demultiplexed and quantified with Cell Ranger or STARsolo (mapping reads to cell barcodes and UMIs) → quality filtering (removing empty droplets, dead cells with high mitochondrial reads) → normalisation (scran pooling or sctransform) → dimensionality reduction (PCA → UMAP or tSNE) → clustering (Seurat Louvain, Scanpy Leiden) → cell type annotation (marker gene expression, reference dataset transfer with SingleR or Seurat label transfer) → differential expression between conditions (pseudobulk with DESeq2, accounting for individual samples as biological replicates) → trajectory analysis (pseudotime: Monocle3, scVelo RNA velocity) identifying developmental trajectories and lineage relationships from transcriptional data.

Example 4: Metagenomics Analysis

Shotgun metagenomics analysis of microbiome samples: raw reads host-decontaminated (bowtie2 mapping to host genome, removing human reads) → species-level classification (Kraken2 k-mer classification against curated NCBI database, Metaphlan3 using clade-specific marker genes for relative abundance profiling) → functional profiling (HUMAnN3 mapping to UniRef90 protein families) → de novo assembly (MEGAHIT, metaSPAdes) → binning (MetaBat2, MaxBin2 separating assembled contigs into genome bins by coverage + tetranucleotide composition) → bin quality assessment (CheckM evaluating completeness and contamination via single-copy marker genes) → taxonomic classification of bins (GTDB-Tk) → metagenome-assembled genomes (MAGs) for novel species discovery from uncultured microorganisms.

Example 5: ChIP-seq and ATAC-seq Analysis

ChIP-seq identifies genome-wide protein binding sites: chromatin immunoprecipitation enriches DNA bound by a target protein; reads aligned to reference genome; peaks called with MACS2 (model-based analysis using local background normalisation); peaks annotated to nearest gene (ChIPseeker); motif analysis identifies enriched transcription factor binding motifs in peaks (HOMER, MEME-ChIP). ATAC-seq identifies open chromatin (accessible regulatory elements): Tn5 transposase tagmentation inserts sequencing adaptors preferentially in open chromatin; reads aligned reducing mitochondrial reads; MACS2 peak calling at nucleosome-free regions (NFRs, <150 bp) and mononucleosome positions; differential accessibility analysis across samples/conditions identifies regulatory elements activated or silenced. Integration of ChIP-seq, ATAC-seq, RNA-seq, and Hi-C data reconstructs transcriptional regulatory networks.

Example 6: Phylogenetic Analysis Pipelines

Molecular phylogenetics constructs evolutionary trees from sequence data. Workflow for viral/bacterial outbreak genomics (e.g., SARS-CoV-2 lineage surveillance): genome assembly → MAFFT multiple sequence alignment → trimming alignment artifacts → IQ-TREE2 or FastTree maximum likelihood phylogenetic tree construction (GTR+G substitution model selection) → time-scaled Bayesian phylogenetic inference (BEAST2 with TempEst root-to-tip regression confirming clock signal) → lineage classification (Nextclade, Pangolin for SARS-CoV-2 nomenclature) → ancestral reconstruction and spatial spread analysis (TreeTime, BEAST discrete phylogeography). Real-time SARS-CoV-2 genomic surveillance (Nextstrain) generated one of the most detailed real-time pandemic phylogenomic datasets in history, tracking >10 million sequenced genomes.

Example 7: Machine Learning in Bioinformatics

Machine learning and deep learning are transforming bioinformatics: DeepVariant uses convolutional neural networks (CNNs) on pileup images from read alignments achieving higher SNP/indel calling accuracy than GATK in challenging regions. scVI and Scanorama use variational autoencoders for single-cell data integration across batches. Protein language models (ESM-1b, ESM-2, ProTrans) learn evolutionary constraints from millions of sequences—enabling mutation effect prediction (zero-shot fitness prediction), protein engineering, and structure prediction (ESMFold). Graph neural networks on molecular graphs predict drug-target interactions, ADMET properties, and molecular toxicology. Bedtools, deeptools, and gkmSVM use SVMs for genomic feature prediction. Federated learning enables multi-site model training on private clinical genomic datasets without sharing raw data.

Example 8: Nanopore Real-Time Sequencing

Oxford Nanopore Technology (ONT) sequencing using MinION, GridION, or PromethION devices generates real-time reads by detecting disruptions in ionic current as DNA/RNA translocates through protein nanopores. Basecalling (converting current signals to base sequences) is performed by deep learning models (Guppy, Dorado). ONT enables direct RNA sequencing (no reverse transcription, preserving RNA modifications), native base modification detection (m6A, 5mC directly from current signal without bisulfite conversion), and ultra-long reads (>4 Mb in single reads). Clinical applications: outbreak genomics (Ebola sequencing in West Africa using portable MinION), real-time pathogen identification in sepsis (direct metagenomics from clinical samples), and rapid whole genome sequencing in intensive care unit rare disease diagnosis within 6-8 hours from sample to VCF.

Try it live

Everything above runs in your browser — open Sequence Alignment DP Landscape and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.