💻 Bioinformatics: Historical Timeline & Scientific Milestones

Compiled the world's first computerized collection of all 65 known protein sequences in 'Atlas of Protein Sequence and Structure', inventing single-letter amino acid codes to save computer punch-card memory.

Invented the dynamic programming algorithm for optimal global pairwise sequence alignment, finding the absolute highest-scoring end-to-end alignment between two amino acid or nucleotide sequences.

Developed the first computer software suite (including gap4) to store, assemble, and edit overlapping shotgun DNA sequencing reads from Sanger sequencing runs.

Constructed the PAM (Point Accepted Mutation) scoring matrices based on observed amino acid replacement frequencies across 1,572 closely related homologous proteins (>=85% identity).

Established the world's first public centralized nucleotide sequence repository to collect and curate rapidly accumulating DNA sequences published in scientific journals.

Adapted dynamic programming to identify optimal local alignments (conserved protein domains or motifs) between sequences, resetting negative matrix scores to zero.

Founded the national public computerized DNA database GenBank, transitioning in 1992 to the National Center for Biotechnology Information (NCBI) under the National Library of Medicine.

Developed FASTA (and FASTP), the first widely adopted heuristic algorithm capable of searching entire sequence databases thousands of times faster than dynamic programming.

Created Swiss-Prot, a non-redundant, manually annotated protein knowledgebase integrating curated scientific literature, post-translational modifications, 3D structure links, and function.

Established NCBI under the direction of David Lipman to develop automated information systems for molecular biology data, building the Entrez unified cross-database search engine.

Invented the heuristic BLAST algorithm that searches massive sequence databases orders of magnitude faster than FASTA by finding short high-scoring segment pairs (HSPs) without gaps (1990 paper is among the most cited in science).

Derived the BLOSUM (Blocks Substitution Matrix) series by directly counting amino acid substitutions in the BLOCKS database of aligned ungapped protein sequence segments.

Developed CLUSTAL W, incorporating position-specific gap penalties and sequence weighting into progressive multiple sequence alignment algorithms.

Introduced Position-Specific Iterated BLAST (PSI-BLAST), constructing a customized Position-Specific Scoring Matrix (PSSM) profile from significant hits of a BLAST search to iteratively detect remote evolutionary homologues.

Built the HMMER software package, implementing profile Hidden Markov Models (profile HMMs) to probabilistically represent sequence consensus, match, insertion, and deletion states across protein families.

Created a unified, species-independent controlled vocabulary to annotate gene function across all organisms, structured as Directed Acyclic Graphs (DAGs) in three domains: Molecular Function, Biological Process, and Cellular Component.

Launched the Ensembl automatic genome annotation pipeline and interactive web browser to provide comprehensive gene predictions, transcripts, and comparative genomics for the emerging human genome and chordate models.

Developed the high-performance UCSC Genome Browser, displaying aligned tracks of genes, regulatory elements, conservation scores (phyloP, phastCons), and epigenetic modifications across the human genome.

Consolidated Swiss-Prot, TrEMBL, and PIR into UniProt, the world's most comprehensive catalog of protein sequence and functional annotation.

Developed BWA, utilizing the Burrows-Wheeler Transform (BWT) and FM-index data structures to align hundreds of millions of next-generation short sequencing reads to the human genome with low memory (~3 GB RAM).

Established the Sequence Alignment/Map (SAM) tab-delimited text format and its BGZF-compressed indexed binary equivalent (BAM), accompanied by the SAMtools manipulation suite.

Created the GATK Best Practices pipeline for discovering germline single-nucleotide polymorphisms (SNPs) and small indels from high-throughput sequencing data.

Developed DESeq2 in R/Bioconductor, using negative binomial generalized linear models with empirical Bayes shrinkage to estimate dispersion and log2 fold-changes across RNA-seq count datasets.

Created the Seurat R package for quality control, normalization (SCTransform), dimensionality reduction (PCA, UMAP, t-SNE), clustering, and integrated multi-modal analysis of single-cell transcriptomes.

Demonstrated at CASP14 that the AlphaFold2 deep neural network predicts protein 3D atomic structures with crystallographic accuracy directly from amino acid sequences.

Trained a 15-billion-parameter Transformer language model (ESM-2) on millions of unaligned evolutionary protein sequences, predicting atomic 3D structures (ESMFold) in seconds directly from single sequences without computing MSAs.

Trained large Transformer foundation models on tens of millions of single-cell transcriptomes, pre-training on masked gene expression to predict cell-type annotations, perturbation responses, and gene network regulatory dynamics.

Released AlphaFold3, replacing the structural module with a diffusion-based generative model capable of predicting the joint 3D structure of all biomolecular complexes: proteins, DNA, RNA, post-translational modifications, small-molecule drugs, and ions.