CINF 1: Generating Canonical Identifiers For (Glycoproteins And Other Chemically Modified)...

Generating canonical identifiers for (glycoproteins and other

chemically modified) biopolymers

Roger Sayle , john may & Noel O’Boyle

Nextmove software, cambridge, uk

250th ACS National Meeting, Boston, MA. Sunday 16th August 2015

motivation

• Non-standard peptides, post-translationally modified proteins and drug-antibody conjugates are becoming increasingly relevant to the life sciences.

• Registration of biologics, beyond the FASTA sequence, is considered desirable but technically challenging.

• In this talk, I discuss complementary approaches to biologics registration; one based upon expressive all-atom representations, another on tracking deltas to a reference database of protein sequences.


Real world small scale example

• Many research reagents contain “hybrid molecules”

– Innovagen SP-5125 lauroyl-apelin-13 • dodecanoyl-QRPRLSHKGPMPF

– Innovagen SP-5126 myristoyl-apelin-13 • tetadecanoyl-QRPRLSHKGPMPF

– Innovagen SP-5124 palmitoyl-apelin-13 • hexadecanoyl-QRPRLSHKGPMPF

– Innovagen SP-5127 steroyl-apelin-13 • octadecanoyl-QRPRLSHKGPMPF


The cutting edge of biosimilarity

• The high prevalence of potentially life-threatening hypersensitivity reactions to the antibody cetuximab (Erbitux) in some US states has been traced to its glycosylation [containing a Gal(a1-3)Gal epitope].

Chung et al., “Cetuximab-induced anaphylaxis and IgE specific for galactose-alpha-1,3-galactose”, New England Journal of Medicine, Vol. 358, No. 11, pp. 1109-1117, 13th March 2008.

• Similarly, Human Erythropoietin (EPO) alpha, beta, delta and omega share the same primary sequence, but differ in their glycosylation patterns.


Monomer dictionaries don’t scale

• Systems based upon monomer dictionaries (such as HELM and PDB) are notoriously difficult to maintain.

• The limited number of monomers in proteinogenic peptides and natural nucleic acid sequences leads to a false sense of security; that monomers are finite.

• In practice, the number of monomers, post-translational and chemical modifications is infinite.

• Even more difficult than standardizing monomer definitions via a central repository, like PDB, is allowing local custom definitions.


48 hexopyranoses


264 deoxy-hexopyranoses

9540 substituted hexopyranoses (4 most common

substituents)

The current situation

• Pistoia HELM can’t [yet] handle/canonicalize glycans and oligosaccharides.

– It can’t uniquely canonicalize Fmoc-Ala-OH (between Pistoia and ChEMBL monomer sets).

• IUPAC InChI can’t [yet] officially handle more than 1024 atoms.

• Folks working on glycoproteins are screwed…

(or use expensive commercial software)


Constructive suggestion…

• Ideally, a chemical identifier should be independent of the input representation or file format.

• Equivalence between small molecules, peptide and proteins are best determined by a single identifier, preferably the existing standard InChI.

• This is possible as increases in computer power and storage mean that cheminformatics toolkits can handle huge biopolymers on modern hardware.


Three recent experiments

1. Is it possible to generate standard InChI for extremely large molecules (polymers)?

2. How well do all-atom canonicalization algorithms scale and can they be improved?

3. Are there alternative canonical identifiers that can be useful in bioinformatics and precision medicine?


Previous inchi key record [(2014)

• Sequence Identifier: UTP10_KLULA

• Sequence Length: 1774 amino acids

• Molecule size: 28509 atoms

• InChI Length: 119699 characters

• InChI key: PHBRSEQMAKHFGD-ZBXWIJJNSA-N

• InChI Canonicalization Time: 73.2s

• Canonical SMILES Length: 35408 chars

• OEChem SMILES Canonicalization Time: 0.4s


Special classes of molecules

• Alkanes – InChI=1S/C6H14/c1-3-5-6-4-2/h3-6H2,1-2H3

– InChI=1S/C8H18/c1-3-5-7-8-6-4-2/h3-8H2,1-2H3

– InChI=1S/C10H22/c1-3-5-7-9-10-8-6-4-2/h3-10H2,1-2H3

– 1 million carbons, InChI is 6,889,942 bytes (~6.9Mbytes)

– 1 billion carbons, InChI is 9,888,888,954 bytes (~9.9Gbytes)

• Polyalanine – InChI=1S/C3H7NO2/c1-2(4)3(5)6/h2H,4H2,1H3,(H,5,6)/t2-/m0/s1

– 1 thousand L-alanines, InChI is 42,965 bytes (~43Kbytes)

– 1 million L-alanines, InChI is 66,888,995 bytes (~66.9Mbytes)

• Theoretically one could write an efficient fasta2inchi


Algorithm scaling to 100AA


Algorithm scaling to maximum


Spread of algorithm run-times


peptide names (chembl)

• The following names are machine generated

• [15-L-arginine]nociceptin CHEMBL526333

• [2-4-chloro-L-phenylalanine]neuropeptide S [human] CHEMBL441576

• [1-L-threonine]cyclosporin A CHEMBL2370014

• [6-L-tryptophan]sermorelin free acid CHEMBL440438

• angiotensin II (3-8) CHEMBL261120

• nociceptin amide CHEMBL389521

• acetyl-alpha-MSH (4-10) amide CHEMBL410411

• [2-L-cysteine,13-L-cysteine]neurotensin disulfide CHEMBL3278512

• myristoyl-[1-L-lysine,4-L-tryptophan]tetrapandin 2 amide CHEMBL3288219

• [2-(4RS)-thiazolidine-4-carboxylic acid,4-L-proline]endomorphin-2 CHEMBL126611

• [22-L-serine]kalata B1 CHEMBL1801140


Scaling-up protein variant naming

• The algorithm described for naming peptides can also be applied to naming arbitary protein variants.

• Consider the a database of the following 11 peptides: – CFFQNCPRG phenylpressin

– CFVRNCPTG annetocin

– CFWTSCPIG octopressin

– CYFQNCPRG argipressin

– CYFQNCPKG lypressin

– CYFRNCPIG cephalotocin

– CYIQNCPLG oxytocin

– CYIQNCPPG prol-oxytocin

– CYIQNCPRG vasotocin

– CYIQSCPIG seritocin

– CYISNCPIG isotocin


Dag representation of sequences


These 11 peptides may be efficiently represented and search as a “directed acyclic graph” [38 vs. 99 states]

entirety of uniprot/swissprot

• Using this representation, all 540546 protein sequences in uniprot_sprot, which contains over 192M amino acids, requires 142M states (1.4Gb).

• This data structure allows close analogues to be identified much faster than using NCBI blastp.

• For example, all 540546 sequences can be queried against this database (i.e. all-against-all) in ~9m30s on a single core on a laptop.

• The sequence from PDB 1CRN (crambin 46AA) is canonically named as [L25I]P01542 in 0.002s.


Application to precision medicine

• A more realistic example is that sequence of the gene “spastic paraplegia4” with six mutations from OMIM:604277 can be canonically named as [I344K,S362C,N386S,D441G,C448Y,R499C]Q9UBP0

• Run-time for this query is 0.2s.

• By comparison, blastp 2.2.29+ takes about 6s.

– With default arguments, NCBI blastp run time is 7s.

– Only 6s with –num_descriptions 1 –num_alignments 1.


conclusions

• “InChI for large molecules” can be achieved, and remain compatible with small molecule InChI identifiers, through the evolution of ever better canonicalization algorithms.

• Journal reviewers who claim that the run-time of canonicalization algorithms is a non-issue, and not an area ripe for improvement are… very mistaken.


acknowledgements

• Greg Landrum, Novatis, Basel, Switzerland.

• Nadine Schneider, Novartis, Basel, Switzerland.

• Evan Bolton, NCBI PubChem project, Bethesda, MD.

• Joann Prescott-Roy, Novartis, Cambridge, MA, USA.

• Daniel Lowe, NextMove Software, Cambridge, UK.


CINF 1: Generating Canonical Identifiers For (Glycoproteins And Other Chemically Modified)...

Science

Transcript of CINF 1: Generating Canonical Identifiers For (Glycoproteins And Other Chemically Modified)...