🧬 Protein Conservation Explorer

Live app: hemoglobin-explorer.streamlit.app

A Python and Biopython project that downloads the same protein from many species, lines them up, and shows which parts evolution has kept the same, as a chart, a family tree and a spinnable 3D model.

It started as a set of scripts studying hemoglobin beta across six mammals and grew into an interactive web app that works for almost any gene, any protein structure, and a set of wood-digesting mushroom enzymes.

What the app does

The app has three sections:

Section	What it's for
🔬 Structure viewer	Search any protein in UniProt and see it in 3D, with its active sites marked
🧬 Gene explorer	Compare one gene across up to 34 vertebrates
🍄 Mushroom enzymes	Compare one enzyme across 18 mushrooms and other fungi

🔬 Structure viewer

Search all of UniProt by name (laccase), gene (TP53), enzyme number (1.10.3.2) or accession (P68871), optionally narrowed to one species. Both boxes suggest matches in a dropdown as you type. Pick a result (✅ marks entries reviewed by UniProt's curators) to see:

• AlphaFold's predicted 3D structure, colored as a rainbow from start to end or by AlphaFold's own confidence. If AlphaFold has no model, the protein's experimental structure from the Protein Data Bank is shown instead.
• Active sites and binding sites from UniProt, drawn right on the model (magenta for active sites, yellow for binding sites). UniProt and AlphaFold number residues the same way, so the marks line up exactly.
• The protein's facts: species, gene, enzyme number, length, what it does, and a table of every site.

A 4-character PDB ID (like 4HHB) opens that experimental structure directly.

🧬 Gene explorer

Pick a gene (type a symbol like TP53 or a word like insulin) and a set of species, from the original six mammals up to every vertebrate in the list. The app then:

1. Finds the protein in each species on NCBI, trying three ways: by gene name, through NCBI's ortholog lists (which link the same gene across species even when its name differs, like mouse Hbb-bs), and by protein name.
2. Aligns the sequences with FAMSA, a fast multiple sequence aligner.
3. Scores every position by the fraction of species that share the same amino acid.

Results are shown in six tabs:

Tab	What it shows
🌳 Family tree	An interactive tree built from the alignment (neighbor joining or UPGMA)
📊 Conservation	Bar chart of how conserved each position is; hover a bar to see every species' amino acid
🧊 3D model	The human protein in 3D, colored by conservation. Hemoglobin uses its real X-ray structure (PDB 4HHB) with the heme groups and key sites labeled; other genes use AlphaFold's predicted structure with UniProt's active and binding sites marked
🐾 Species	A short note on each species, how long ago it split from humans, and its percent identity to human
🧬 Alignment	The full alignment
📄 Sources	Every NCBI record used, with links, protein names, lengths, and how each one was found

Everything can be downloaded: the alignment (FASTA), conservation scores (CSV), the tree (Newick), and any chart as a PNG.

🍄 Mushroom enzymes

The same analysis for fungi. Pick an enzyme and a set of fungi, from button mushrooms and shiitake to turkey tail, reishi, dry rot and baker's yeast. Enzymes include:

• Laccase, manganese peroxidase and lignin peroxidase, which break down lignin, the tough glue in wood (used for dye removal, bioremediation and pulp bleaching)
• Cellobiohydrolase, which breaks cellulose into sugar (used in biofuel enzyme mixes)
• Tyrosinase, which makes melanin and is why cut mushrooms turn brown
• Chitin synthase, which builds fungal cell walls and is a target for antifungal drugs
• Actin, a highly conserved control to compare the enzymes against
• Or type any other enzyme name or enzyme (EC) number

For each fungus, the app:

1. Searches UniProt by enzyme number and name, full-length proteins only, skipping look-alikes such as "actin-binding protein" when you asked for actin.
2. Picks the matching copy. Fungi often carry several copies of an enzyme, and two different copies can be less alike than the same copy in two species (some mushrooms have a second, unusual actin that is only about 56% identical to the normal one). The app compares each fungus's top candidates to a curator-reviewed reference and keeps the closest, so the alignment compares like with like.
3. Falls back to NCBI (RefSeq, then GenBank) for fungi UniProt has nothing for.

Results show the same tabs as the gene explorer: family tree, conservation chart, 3D model (the AlphaFold model of that exact UniProt entry, with its sites marked), a note on each fungus, the alignment and the sources, each linked to UniProt or NCBI.

Key findings (hemoglobin beta)

• 106 of 147 positions (72%) are identical across human, chimpanzee, rhesus macaque, cow, pig and horse.
• The most important sites never change. Glu6 (the position mutated in sickle cell disease) and the two histidines that hold the heme iron (His63 and His92) are identical in all six species.
• Human hemoglobin alpha and beta are about 42% identical (63 of 149 aligned positions). The two chains came from one ancestral gene that was duplicated and then diverged.

How it works

UniProt / NCBI  ->  FAMSA alignment  ->  conservation scores  ->  chart, tree, 3D model

• Retrieval: Biopython Entrez searches NCBI's RefSeq proteins. Orthologs are found with NCBI Gene's ortholog groups and linked to their RefSeq proteins. Fungal enzymes and the structure viewer use the UniProt REST API.
• Alignment: pyfamsa (FAMSA) for the multiple alignment; Biopython PairwiseAligner for pairwise comparisons and for picking the matching copy of an enzyme.
• Tree: Biopython Phylo distance methods (neighbor joining and UPGMA).
• 3D: structures from the RCSB Protein Data Bank and the AlphaFold database (looked up through UniProt), lined up with the alignment and drawn with py3Dmol.
• App: Streamlit, with Plotly for the interactive charts.

Files

File	What it does
app.py	The Streamlit web app and the gene explorer
analysis.py	Alignment, conservation scoring and sequence similarity
databases.py	UniProt searches, active sites and AlphaFold models
structure_viewer.py	The Structure viewer section
mushrooms.py	The Mushroom enzymes section, with notes on each fungus and enzyme
structure3d.py	Downloads 3D structures and colors them by conservation
style.py	Page styling and the header
species_bios.py	Notes and icons for each vertebrate species
species_list.py, gene_list.py	Species, fungi and gene suggestion lists
tree.py	Builds and draws the family tree
read_fasta.py	Reads a FASTA file with SeqIO
compare.py	Pairwise alignment of human hemoglobin alpha vs beta
fetch.py, fetch_species.py	Download proteins from NCBI
align_all.py	Multiple alignment with FAMSA
conservation.py	The original conservation chart with the key hemoglobin sites marked
tests/	Automated tests, run on every push by GitHub Actions

Run it yourself

git clone https://github.com/evanyoung1022-oss/Bio-Python-sequence-alignment.git
cd Bio-Python-sequence-alignment
pip install -r requirements.txt
streamlit run app.py

NCBI asks every user of its API for an email address. Put yours in .streamlit/secrets.toml as NCBI_EMAIL = "you@example.com". An optional free NCBI_API_KEY there makes NCBI searches about 3x faster.

To run the tests:

pip install pytest
pytest

The tests check alignment, conservation scoring, tree building (including the case where every sequence is identical), how UniProt searches are built and read, picking the matching enzyme copy, and that the whole app starts without errors.

Limitations

• Results depend on how each species is annotated in UniProt and NCBI. Some species have no record for a protein, so they are skipped.
• In the gene explorer, when a species has several versions (isoforms) of a protein, the app uses the first one NCBI returns.
• Picking the matching enzyme copy works by similarity, so for enzymes with many very different copies (chitin synthase comes in several classes), the fungi may still be compared across copies.
• Many fungal UniProt entries are unreviewed and have no active sites listed, so their 3D models may show no site marks.
• A gene tree shows the history of one protein, which doesn't always match the history of the species.
• AlphaFold structures are predictions, and regions it isn't confident about often look like loose loops.

Built with

Python, Biopython, NCBI E-utilities, UniProt REST API, FAMSA (pyfamsa), Streamlit, Plotly, py3Dmol, RCSB PDB, AlphaFold DB, matplotlib, pytest, GitHub Actions

About

Built by Evan Young, studying Integrated Science and Technology (biotechnology) at James Madison University. Released under the MIT License.