Protein informatics and protein data

Protein informatics uses computers to learn about proteins: where their functional sites are, what they do (their biochemical and biological functions), and even the 3-D shapes of hypothetical proteins: proteins predicted from a genome sequence but never actually detected, whose job is unknown. What once took weeks in a lab, a computer can often suggest in minutes, from the sequence alone.

Protein data comes in many forms (raw data):

• Microscope images of clumps (aggregates) of heat-damaged protein.
• Measurements of proteins in solution.
• Short sequence fragments from MALDI mass spectrometry, and the full sequences assembled from them.
• Crystal structures, stored as PDB files.
• Files of interactions between proteins.
• NMR and mass spectrometry data.
• Hypothetical proteins predicted from genomes.

Each has its use: aggregate images help design protein markers; solution data give physical, chemical and speed (kinetic) properties; MALDI fragments are used to rebuild the full sequence; crystal structures show mutations and interactions; and maps of protein networks point to possible drug targets.

Two things are needed: raw data from databases (NCBI, PDB, ChEMBL, BioModels) and analysis tools: wavelet image analysis, similarity and homology searches, structure optimisation, machine learning (ANN, artificial neural networks; SVM, support vector machines; HMM, hidden Markov models), network mapping, and SBML (a language for describing biological models).

The great advantage of computer prediction: it works even when only the gene sequence is known, and it is quick, cheap and can handle huge numbers of proteins at once (high-throughput).