Why do we need seondary structure detection algorithms?

Question

There are many secondary structure prediction and assignment algorithms (e.g., DSSP, STRIDE, etc.) available.

On the other hand, I see that PDB files have secondary structure information available in them.

For example: the following are the lines from the PDB file 6jue.pdb:

HELIX    1 AA1 GLY L   53  GLY L   59  1                                   7    
HELIX    2 AA2 ALA L   79  SER L   92  1                                  14    
SHEET    1 AA1 4 GLU L  10  LEU L  18  0                                        
SHEET    2 AA1 4 GLY L  99  ARG L 107 -1  O  ILE L 101   N  VAL L  16           
SHEET    3 AA1 4 GLN L  66  VAL L  70 -1  N  ILE L  68   O  ILE L 104           
SHEET    4 AA1 4 GLU L  73  SER L  74 -1  O  GLU L  73   N  VAL L  70           
SHEET    1 AA2 3 ALA L  40  ILE L  49  0                                        
SHEET    2 AA2 3 VAL L  28  SER L  35 -1  N  ASN L  33   O  LEU L  42           
SHEET    3 AA2 3 ILE A  87  THR A  89 -1  O  ILE A  88   N  VAL L  30           

So, my question is, if PDB files already supply secondary structure information, why do we need DSSP and STRIDE-like algorithms?

Answer 1

Right, that's easy PDB is strictly experimentally derived structures, for example crystal structures or NMR. There are at least three very good reasons,

1. There are billions of proteins. Even small differences in amino acid homology can have a drastic shift in function via structural changes. It is simply not possible to obtain experimental data for every protein, thus predictions are required in this case secondary structure or e.g. AlphaFold.
2. Experimentally derivations are static but real world proteins are dynamic and that is lost in PDB data. Some surface structures are so dynamic they can't be viewed e.g. via NNR. Secondary structure prediction has some value here because these structures are highly hydrophilic.
3. Certain proteins can't be viewed experimentally, for example transmembrane proteins. Secondary structures are essential here because alpha-helices are a central unit for transmembrane proteins and can be reliably predicted by most methods.

---

Responding to the comment. Amino acid homology definition A homologue is the identical amino acid position between the same protein between two or more genomes. So if the common ancestor between the two genomes introduced a mutation at this position, both ancestral genomes will carry the mutation. For example the haemoglobin genes in mammals (possibly all vertebrates), the pattern of mutation precisely follows the evolution of the species. This contrasts to an orthologue, where two proteins with high percentage identity don't belong to the exact same protein in the same common ancestor, for example they have functional differences and evolved independently. For example, HLA complex (immune superfamily proteins) are very similar to each other but have distinct functions in the targeted immune response and evolve "separately" of each other (there are complex caveats which I'll avoid explaining).

Two examples,

A. Sickle-cell anaemia is a single amino acid change resulting in a radical shift in red blood cell tertiary structure. The blood cells start aggregating and shifting conformation from "round" to "sickle" shape. This structure has a reduced capacity to bind oxygen, ie. its function changes.

B. Small molecule binding (i.e. drugs) to pathogen proteins at their active sites. The pathogen must stop the drug from binding by mutation otherwise its lethal and that mutation process ain't trivial:

1. Small amino acid substitutions The thermodynamic stability of proteins is weird, it's called compensatory mutation. It's a massive bank of theory in protein modelling, it's as big as "point mutation" modelling in phylogenetics. Introduce one mutation to a protein, e.g. in the active site and the electrostatic stability of the active site gets "blown apart",i.e. no longer functions because the structure has changed. Introducing one or two "compensatory mutations" in the same active site, can stabilize its protein structure by stabilising the electrostatic charge between the proximal amino acids residues of the secondary/tertiary of the active site resulting in regaining its functionality. This is the entire theoretical basis behind Google DeepMind's "Alphafold", along loads of other algorithms.

Point 1 summary: 1 amino acid difference, massive change in function, 2 or 3 amino acid substitutions the functionality is regained. Again the strict homology and functionality ain't a linear relationship. Just to stress this mutation process must occur all the time because large chunks of theoretical modelling of protein structure are based on this premise, i.e. compensatory mutations.

Other examples between distantly related proteins,

2. Deep evolution In "tree of life" analytics investigators assume that e.g. histones in yeast and the protein homologues in the protists (basal unicellular eukaryotes) have identical functions. I've seen these assumptions in real world research. We've no idea that the homology between two distant species relates to the same function. This field of "deep evolution" is trying to establish the evolution of life from the first common ancestor, but for it to work the proteins must carry the same function in different species. Typically we're talking approximately 60% homology.
3. Low homology, identical function However, distant homologies (percentage similarity) in for example viruses, can carry exactly the same function. For example, the envelope/spike protein in an RNA virus is often around 40% identity to another member of the species in that family. Yet both proteins have the same structural confirmation and the exactly the same function (their host cell specificity can be different).

Points 2 and 3 the homology, e.g. 60% versus 40% has no relationship to the function. Protein structural modelling can and do identify functional similarities much better than strict percentage identity.