Bioinformatics for chemists

Sequence databases, alignment (BLAST), and predicting protein structure and function from sequence.

Undergrad

In a nutshell

Bioinformatics is the use of computers to store, search and make sense of biological data, especially DNA and protein sequences. Modern sequencing machines produce enormous amounts of data, far too much for anyone to read by eye, so scientists rely on public databases and clever search tools. If you discover a new gene, you can paste its sequence into a tool called BLAST and, within seconds, find the most similar genes in every organism ever sequenced. Similar sequences usually mean a shared ancestor and often a similar job, so this is a powerful way to guess what an unknown protein does. Today, programs such as AlphaFold can even predict a protein's three-dimensional shape from its sequence alone, which helps chemists design drugs and enzymes.

Databases and formats

Sequence data are deposited in public repositories that exchange data daily: nucleotide collections (GenBank, the European Nucleotide Archive and the DNA Data Bank of Japan), UniProt for protein sequences with curated annotations of function, modifications and variants, and the Protein Data Bank (PDB) for experimental 3D structures. The simplest exchange format is FASTA: a header line beginning with ">" followed by the sequence in one-letter codes. Annotated records add features such as coding regions, exons and the source organism.

Similarity, identity and homology

Identity: the percentage of aligned positions with the same residue.
Similarity (for proteins): also counts conservative substitutions, such as Leu for Ile.
Homology: shared evolutionary ancestry. It is a yes/no inference, not a percentage; high similarity is the evidence for it. Homologues in different species that diverged by speciation are orthologues; those from gene duplication are paralogues.

Alignment

Aligning two sequences means inserting gaps so that equivalent residues line up, then scoring. Protein alignments use substitution matrices (BLOSUM62, PAM) derived from observed substitutions in related proteins: identical and chemically similar pairs score positively, dissimilar ones negatively, with penalties for opening and extending gaps. Dynamic programming finds the optimal global alignment over full length (Needleman–Wunsch) or the best local region (Smith–Waterman).

BLAST (Basic Local Alignment Search Tool) speeds up database searches heuristically: it finds short exact or high-scoring "words" shared with database sequences and extends them into local alignments. Each hit has a bit score and an E-value, the number of hits of at least that score expected by chance in a database of that size; an E-value of 10^{-30} is overwhelming evidence of homology, while values near 1 are not. Variants include blastn (nucleotide), blastp (protein) and tblastn (protein query against translated DNA). Multiple sequence alignment (e.g. Clustal Omega, MUSCLE) of a protein family highlights conserved residues, often catalytic or structural, and is the basis of phylogenetic trees.

From sequence to structure and function

ProtParam-type tools estimate molecular mass, pI and the extinction coefficient at 280 nm from composition, useful for purification planning.
Domain and motif databases (Pfam, InterPro, PROSITE) assign families and functional sites.
Secondary-structure, transmembrane-helix and signal-peptide predictors flag features from local sequence patterns.
Homology modelling builds a 3D model on a related known structure; deep-learning predictors such as AlphaFold use multiple sequence alignments and learned geometry to predict structures with per-residue confidence scores (pLDDT), and the AlphaFold Database holds predictions for most known proteins.

Predictions are hypotheses: a low-confidence region may be disordered, and inferred function should be checked experimentally.

Key points

GenBank/ENA/DDBJ store nucleotide data, UniProt proteins, PDB 3D structures; FASTA is the simplest format.
Identity and similarity are measured percentages; homology is an inference of common ancestry.
Substitution matrices such as BLOSUM62 and gap penalties score alignments; global vs local alignment.
BLAST finds local alignments fast; the E-value estimates how many hits that good arise by chance.
Structure prediction (homology modelling, AlphaFold) gives models with confidence scores that need experimental testing.

Check yourself

5 questions

1Which database is the main resource for experimentally determined 3D protein structures?
The PDB archives atomic coordinates from crystallography, NMR and cryo-EM.
2Two proteins share 85% sequence identity. What is the most reasonable conclusion?
Homology is all-or-nothing; high identity is strong evidence of common ancestry, but not a percentage of homology.
3What does a very small BLAST E-value (e.g. 10^{-40}) indicate?
The E-value is the expected number of chance hits at that score; tiny values signal a real relationship.
4Why do protein alignments use matrices like BLOSUM62 rather than simple identical/non-identical scoring?
Substitution matrices reflect observed exchange frequencies, improving sensitivity for distant homologues.
5An AlphaFold model has a long region with very low pLDDT. What is the most likely interpretation?
Low per-residue confidence often coincides with disorder and should not be read as a defined structure.

Recent preprints on this topic

Loading…

Matched by keyword from ChemRxiv via Crossref. Not peer-reviewed; read critically.