UniProt makeup - UniProtKB vs UniRef vs UniParc @thebumblingbiochemist

Uploaded August 2026 | Updated September 2026, 2 weeks ago

Information about protein sequences (& their corresponding proteins) is housed in UniProt.

UniProt consists of...
- UniProtKB: UniProt KnowledgeBase. It includes annotated entries-one per protein-encoding gene.
- Further split into:
- Swiss-Prot, which has reviewed data
- TrEMBL, which as unreviewed data
- Swiss-Prot (sp)- reviewed, manually curated (more reliable, more info, often something people have actually experimented with a lab)
- TrEMBL (Translated EMBL Nucleotide Sequence Data Library)(tr)- unreviewed, autocurated (less known, be more cautious, most stuff from less-studied organisms is here)

- UniRef: UniProt Reference Clusters. It includes data clustered by % sequence identity, then represented by a reference entry.
- Further split into:
- UniRef100: 100% identity
- UniRef 90 90% identity - 58% reduction in database size
- UniRef 50 50% identity - 79% reduction in database size
- Helps you search protein sequence data more efficiently
- Clusters are giving accession codes containing "UniRef100”, "UniRef90_", or "UniRef100_", followed by the code of the reference entry.

- UniParc: UniProt Archives. It includes ALL non-redundant protein sequence data. Data in UniParc is the least reliable, but it ensures everything is stored permanently somewhere.

Much more here: ebi.ac.uk/training/online/courses/uniprot-quick-tour/the-uniprot-databases

If you're looking for information about a specific protein, you probably want to be in UniProtKB, which holds annotated entries. Each entry contains information about all versions (isoforms and variants) of the protein made from a single gene. The amount of annotation varies from entry to entry, and is reflected in the annotation score.

Pro Tip: You can BLAST against these databases to quickly find a wider variety of sequences similar to a query sequence.

More on annotations here: thebumblingbiochemist.com/glossary/annotation & youtu.be/z2vyZ6bfGkg

More on databases and when & how to use them here: bit.ly/databases_guide & youtube.com/playlist?list=PLfHM6aSvn5eI

More on Uniprot: bit.ly/uniprotprotparam

More on other tools for working with proteins: thebumblingbiochemist.com/365-days-of-science/proteintools

more about all sorts of things: #365DaysOfScience All (with topics listed) 👉 bit.ly/2OllAB0 or search blog: thebumblingbiochemist.com

#scicomm #biochemistry #molecularbiology #biology #sciencelife #science #realtimechem

UniProt makeup - UniProtKB vs UniRef vs UniParc

Leucine quick hits: ketogenic amino acids; #20daysofaminoacids 2025 Day 5

Tips for using Hamilton syringes & other needle/syringe options for injecting into an FPLC

Annotations of sequence data (proteins, chromosomes, genes, etc.), especially UniProt

$Overview & practical guide to x-ray crystallography maps (2fo-fc, fo-fc (difference map), omit maps) X-ray crystallography maps (viewing & understanding 2Fo-Fc, Fo-Fc, etc.)… In X-ray crystallography, electrons in a crystal interact with x-rays to generate a diffraction pattern. Then crystallographers work backwards from the diffraction patterns to create an electron density map. Then they build an atomic model into that map. blog: https://bit.ly/xraymaps Sounds straight forward right? Wrong! There’s a big problem. Recreating an electron density map from diffraction data basically involves reconstructing a complicated wave through a math-y thing called a Fourier transformation. And a wave equation describing that wave requires both intensity (which we can measure from spot strength) & phases. We only can detect the intensity. Leaving us with a big “phase problem.” So in order to actually create a typical map - even a pretty bad one - we have to guess the phases through various methods like molecular replacement (which is just computational, basing it on similarity to a known structure) or experimental determination with techniques like SAD and MAD or heavy metal soaking that give us only rough starting points. Scientists then compute their best guess models based on the initial data & calculate what the map would look like if the model were correct. Then they refine their models to make them match the data better, working back and forth between map and model. And maps help them find their way! There are a few kinds… 2Fo-Fc is a common way to view electron density. This map shows 2X observed signal (Fo) - modeled (calculated) signal (Fc) (so it weighs observed more highly) & you want to see continuous density. Density is typically clearest in the backbone and stiff areas of a protein and less clear (maybe even invisible) in side chains and flexible areas. These areas have a high “B factor” aka displacement factor. You might see variations of this such as 2mFo-DFc which add some statistical weighting stuff to make more reliable, higher strength, data count more. Fo-Fc shows you where things should or shouldnt be! The Fo-Fc difference map highlights differences between the observed & the calculated-from-the-model signals. So, if they match completely, youll see nothing! but they wont (at least not everywhere), so.… its typically displayed in red/green on top of the 2Fo-Fc map. A negative (red) blob tells you youve modeled in something extra thats not supported by the data. A positive (green) blob tells you the model is missing something - theres still signal you need to account for Note: there will have to be an equal amount of red and green in the map so lots of it will just be noise - the 2Fo-Fc can help you sort it out by tracing the continuous density There are also “omit maps” where you calculate maps based on structures where you’ve left out (omitted) some part of the model (such as a bound ligand (binding partner)). This is often used to show that the density supporting the ligand in your other maps is real and not just the result of model bias. This is really important because scientists often have wishful thinking that something is bound so they model it on which then contributes to the normal map - but not the omit one. So there better be density still there! How much density you see on the screen depends in part on how you’ve set the contour level. Maps can be contoured at different levels and changing the sigma (o) level only changes how much of the data is displayed, not the data itself. o refers to the standard deviation of the strength of the signal (much of which is just noise) above the average, & you choose a factor of this at which to cut off the data. A higher o cut-off means that you only show data whose signal is further above the level of noise, but you also leave out true but weaker signal. Higher o contour is stricter. You just see the strongest signal. It’s important not to set your contour too low when building your model or you will end up building into a bunch of noise! Finished in comments$

The CCP4 bulletin board is the place to be if youre interested in x-ray crystallography

A practical guide to FASTA format for protein & nucleic acid sequences & UniProt FASTA headers