Biotechnology · Intermediate

Proteins Practical

Written by SignalsLab

Early edition: these lessons are being reviewed.

Content

Summary

This lesson covers the computational identification of protein domains using HMM profiles from the Pfam database and the prediction of secondary structure elements using the jnet tool. It emphasizes the importance of indexing models with hmmpress, interpreting fuzzy domain boundaries and e-values, and the careful processing of multiple sequence alignments to avoid information loss.

Intuition

Proteins are not just random chains of amino acids but are composed of distinct, evolutionarily conserved units called domains. To find these, we treat them like patterns in a library, using Hidden Markov Models (HMMs). However, these patterns are not rigid; domain boundaries are often 'fuzzy,' meaning the start and end positions can vary slightly between different protein hits. When predicting secondary structure, we look for local shapes like alpha-helices and beta-strands. While using multiple sequence alignments (MSAs) can improve accuracy by leveraging evolutionary data, we must be careful: removing gaps to clean the alignment can sometimes destroy the very information we need if the sequences are too different.

Derivation

  1. Acquire domain models by downloading Pfam-A.hmm.gz from the Pfam FTP server.
  2. Decompress the file using gunzip to create Pfam-A.hmm.
  3. Index the raw model file using hmmpress to prepare it for efficient searching.
  4. Run hmmscan on a target sequence file (e.g., pdz.fasta) against the indexed library.
  5. Use the --domtblout flag to capture detailed domain hit information.
  6. Interpret results by checking 'ali coord', 'from', and 'to' for domain positions.
  7. Assess statistical significance using the e-value, where lower values indicate better matches.
  8. Predict secondary structure using jnet -p on a FASTA file.
  9. Identify 'H' for alpha-helices and 'E' for beta-strands in the output.
  10. For improved prediction, generate an MSA using tools like clustalw2.
  11. Remove gaps column-wise using trimal with -nogaps to create a gap-free FASTA.
  12. Evaluate if the gap-free alignment retains sufficient biological information.

Worked example

Problem

You have a multiple sequence alignment of PDZ domains generated by clustalw2. You wish to run jnet to predict secondary structure. What is the critical preprocessing step required before running jnet, and what is a potential risk of this step?

  1. Identify that jnet requires a gap-free input for optimal performance in this context.
  2. Use the trimal program with flags -in, -fasta, and -nogaps to remove gaps column-wise.
  3. Recognize that if the sequences are too divergent, removing gaps may render the alignment uninformative.
  4. Conclude that the alignment must be evaluated to ensure it retains sufficient information for accurate prediction.

Answer Remove gaps column-wise using trimal; the risk is that excessive removal may make the alignment uninformative if sequences are too divergent.

Common mistakes

  • Assuming domain boundaries always start at the same fixed position across all hits.
  • Running jnet directly on a multiple sequence alignment without removing gaps.
  • Removing gaps from an MSA without checking if the sequences are too divergent, which can render the alignment uninformative.
  • Misinterpreting the e-value as a probability rather than the expected number of random occurrences.

Animation

Animation coming soon

The animated version of this lesson is being prepared for the web. In the Signals AI app, lessons are drawn step by step.

Code

This is a text lesson: it has no animation code yet.

ccBYNC© 2026 SignalsLab. Lesson text shared as is under CC BY-NC 4.0.