iiinova — where your digital self keeps evolving

Interestingly, even though this demonstrates that the low designability of AFDB is essentially an artefact from RMSD calcs after alignment, using more sophisticated aligners (TMalgin, Sheba) do not achieve the same rescue. 16/19

100

Indeed, we show that if you truncate AFDB strucs to their corresponding PDB entry length, you get the same designability. Without a PDB to compare to, you can also rescue via a simple trick: just compute statistical outliers from an initial alignment, realign without them, and use median RMSD 15/19

Realizing that this gap is RMSD, not pLDDT driven, we speculated the cause to be a PDB artefact: in xray strucs, flexible termini often lack density / unstructured parts were truncated to begin with. As a result, the designability of PDB structures gets very high compared to full-length AFDBs 14/19

100

We also question whether the refolding pipeline always is a robust evaluation metric to begin with. As e.g. noted in @moalquraishi.bsky.social's Genie2 paper, the designability of PDB structures is on average much higher than that of the AFDB. Now, we have a pretty good guess why: 13/19

Sequences from design models like ProteinMPNN boost folding success, but for anything beyond medium sized proteins, you almost never can accurately fold in ss mode. 11/19

100

We next looked at a set of literature-reported experimentally tested designs and compared folding models’ ability to act as “oracles”. Again, evo info was detrimental. MSA mode’s poor performance can be obfuscated though, for seqs where the MSA is very shallow / empty. Duh: empty MSA == ss mode 9/19

Worse: If you repeat this experiment starting from sequences designed with ProteinMPNN, the effect is even worse. The reason likely is that ProteinMPNN designs very strongly and unambiguously encode the intended structure, which can make folding models overconfident about their predictions. 8/19

We found that this trend is exacerbated by the availability of evolutionary info to models: AF2 MSA behaves the worst, ESMfold is a bit better, and AF ss is the best (out of bad bunch). As others noted: the signal from these MSAs / pLM embeddings overrules “reason” 7/19

130

We probed this for ESMfold, and for AF2, where two modes can be employed: the default MSA mode (often seen for redesign of native proteins) and single sequence mode (often seen for de novo protein design) First, we checked how good they are at identifying clearly bad designs 5/19

100

Led by @kerlenkorbeld.bsky.social and Seva Viliuga, we started this project under the premise of the field's most common protein design evaluation workflow: the self-consistency pipeline (function->backbone->sequence->evaluation), where folding models are used for the last step. 2/19

120

Max Fürst @maxfus.bsky.social · 05/11/2025

New preprint🚨 Imagine (re)designing a protein via inverse folding. AF2 predicts the designed sequence to a structure with pLDDT 94 & you get 1.8 Å RMSD to the input. Perfect design? What if I told u that the structure has 4 solvent-exposed Trp and 3 Pro where a Gly should be? Why to be wary🧵👇

46524

I set myself a reminder to check this 1 year later & turns out I was mostly wrong🥲 Here search results from on "DNA", last year (orange) vs now (blue) and the indexing issue is only noticeable for the current year ('24), barely an increase for '23 New hypothesis: all (bio?) topics peaked around '21?

Max Fürst @maxfus.bsky.social · 27/10/2025

wouldn't it make much more sense to plot this relative to normal distribution found across the proteome? Too lazy to extract the numbers and make a scatter plot, but I think Ala and Glu are at least as off the axis as the aromatics www.uniprot.org/uniprotkb/st...

010

Max Fürst @maxfus.bsky.social · 05/10/2025

Un-frigging-believable. I have waited for something like this for so long!

24511

Max Fürst @maxfus.bsky.social · 08/08/2025

I kept telling people this in discussions, until someone German insisted so I looked it up and it's indeed not that clear. Here formatted in a similar way from source www.bka.de/DE/AktuelleI...

000

Max Fürst @maxfus.bsky.social · 01/08/2025

Now that OpenCRISPR is in nature and rekindled the 'what's-a-novel-sequence' debate, I'm happy to share an app to check this, which I built for fun some time ago. fuerstlab.shinyapps.io/SeqNovelty/ quick 🧵

11710

Max Fürst @maxfus.bsky.social · 02/06/2025

Fixed that for you! Original (very awkward) color scale did not reflect how much of an outlier NL is. Picture changes for 3yr - school age though. Of course this data does not distinguish between day care vs grandparents etc. Not that the latter is an option for expats, anyway 😶‍🌫️

010

As a final application example, we demonstrate protein immobiliziation. Commercial azido agarose reacted happily with ADD-tagged GFP, and we got a brightly shining resin after reacting the protein on-column with ApbE!

Immobilization of the EGFP model protein onto agarose azide. Covalent attachment is demonstrated by the retention of EGFP on the agarose resin after extensive washing with a buffer containing 350 mM imidazole.

100

Conjugating commercial click-modified DNA oligos also worked at the first shot without optimization, although at relatively low conjugation yields. We’re not so experienced with the chemistry, but apparently Cu catalysis isn’t great for DNA either, so probably switching to SPAAC would work better.

Demonstration of chemoenzymatic protein-DNA conjugation via ADD-tagged proteins. SDS-PAGE analysis reveals the creation of a protein-DNA conjugate, detected by Coomassie staining, SYBR Gold staining, and in-gel fluorescence analysis.

100

As expected, biomolecule conjugation needs more care: we saw moderate conversion yields in protein-protein conjugations (N-N/C-C/branched). We did not optimize our proof of concept—safe to assume that it could easily be pushed by varying position, adding linkers, optimized reaction conditions, etc.

Protein-protein conjugation using CuAAC and ADD-tagged proteins. (a) Conjugation of terminal N-to-N fused proteins using complementary functionalized F1-MBP. (b) Conjugation of terminal C-to-C fused proteins involving alkynyl GFP-F2 and azido GFP-F2. (c) Conjugation of branched-branched proteins with complementary functionalized NqRC. (d) Conjugation of terminal-branched proteins involving NqRC and F1-MBP.

100

Small molecules were readily conjugated to all our example proteins: we show attachment of fluorophores & biotin, and establish that both classic copper-catalyzed, as well as @carolynbertozzi.bskyverified.social's copper-free SPAAC reaction are compatible with ADD tags & result in full labeling.

Small molecule decoration of modified target proteins using click chemistry. a) In-gel fluorescence was used to follow the reaction of alkynyl NqRC with Cy3 azide and showed saturation within 20 min. b) Reaction of alkynyl NqrC with azido-biotin-PEG4, demonstrating quantitative incorporation of biotin-PEG4 as indicated by a complete band shift after 10 minutes. c) Reactivity of N-terminally alkyne-modified MBP with Cy3 azide. d) SPAAC reaction of F1-MBP with DBCO PEG4 alcohol. MS analysis showed full conjugation as indicated by mass addition of 510 Da.

Because ADPRC works on NAD+, not NADPH, while ApbE only accepts NADPH and ignores NAD+, we can even run the two enzymatic steps (substrate generation & protein modification) in one pot—just add the two enzymes with the precursors NAD+ and the click alkanol to your target protein!

One-pot click chemistry handle attachment (alkynyl-phospho-ribose) to F1-MBP using ADPRC and ApbE in Tris·HCl pH 8 starting from NAD+ and 4-pentyn-1-ol. Complete labelling was observed after 12h reaction.

100

Turns out ApbE is happily accepting this substrate, allowing us to label a variety of proteins at N or C terminus, as well as internally, with either alkynyl or azido ribose phosphate moieties, yielding typically full conversion after a few hours of reaction, as determined by MS.

Bioorthogonal handles (alkynyl and azido ADPR) conjugation to various proteins with ADD-tag located C-terminally, N-terminally and at an internal loop. (a) Reaction conditions for the attachment of bioorthogonal handles to target proteins using ApbE. (b) MS analysis showing the successful incorporation of alkynyl and azido ribose phosphate, indicated by mass additions of 281 Da and 296 Da, respectively. "F1" and "F2" denote the tag sequences GVDGLSGATLTS and GVDGLSGATL, respectively.

100

Given the enzyme’s reported promiscuity toward NADH, we hypothesised that it tolerates synthetic substrates with similar structure. Courtesy to Jäschke lab developing NAD capture seq, we knew that a commercial enzyme—ADPRC—can generate a click dinucleotide from cheap small molecule alcohols & NAD+

Conversion of NAD+ to a click dinucleotide following the strategy developed for NAD captureSeq
https://www.nature.com/articles/nature14020

130

🚨 preprint 2️⃣ this month: our (purely experimental🧪) venture into #ChemBio We prouldy present: ADD-tagging of proteins (or "ADDing") —a super convenient enzymatic technique to install click chemistry handles on proteins. Led by superstar @wahyuwidodo.bsky.social www.biorxiv.org/content/10.1... A 🧵👇🏽

Overview of ADD-tagging including ADP-ribosyl cyclase (ADPRC)-catalysed dinucleotide substrate generation and two step chemoenzymatic labelling of target proteins with the flavin transferase ApbE, followed by click chemistry-based functional group attachment.

45316

BoostMut's output is a table of scores on various metrics & we provide a web app to explore. While the default score is simple mean, we trained some lightweight ML models to assigned weights. They did a bit better, we're currently still exploring!

100

We also ran fresh MDs on large literature datasets that used ML predictors: again, small but consistent ROC AUC gains. Finally, we tested BoostMut's top mutations in the lab, and: even higher success rate than manual selected mutations!

100

We made a huge effort to benchmark thoroughly, mostly via ROC analysis. First on old FRESCO projects: predictions & MDs were already there. BoostMut's top hits overlapped manual picks, though not 1:1. But: success rates were as good or better.

100

Built on Python’s MDAnalysis, BoostMut formalizes key stability checks: H-bonds ↑, hydrophobic exposure ↓, flexibility ↓, etc. Some metrics were tricky, e.g. what’s “normal” for sidechain flex? We simulated high-res PDBs, binned by surface exposure, & fit a monotonic curve to not penalize rigidity.

100

Given that this is in the same realm as what we achieved in previous FRESCO projects, e.g. when we engineered an alcohol dehydrogenase to boiling point stability (5 years ago) we consider the stabilizing mutation prediction problem as not solved at this point in time elifesciences.org/articles/54639

100

While primary prediction could these days be ML-based, we haven't really given up on energy-based ones, due to very similar hit rates. Take e.g. PRIME, a recent "SOTA" model. Great paper, as (unlike most new fancy model papers) they actually put their tool to the test in the lab, but hit rate = 40%

https://www.science.org/doi/epdf/10.1126/sciadv.adr2641

100

We've engineered thermostable proteins in Groningen for many years, traditionally using our FRESCO pipeline, which uses MD screens to sort out false positives from thermostability predictors such as FoldX. This has been highly successful, stabilizing proteins on average by +20 °C with ~8 mutations.

100

Preprint from the lab🚨 Have you ever engineered proteins to be more stable and were unhappy about your predictor's success rate? We got you covered with BoostMut! Great work led by @kerlenkorbeld.bsky.social now online at www.biorxiv.org/content/10.1... A thread 🧵

1418

Covid Spike protein, PDB 6CRZ

260

copy your favorite outcome to Adobe Illustrator and use Image Trace with the preset 3 or 6 Colors. Expand, and adjust colors to your liking

160

Screenshot (make sure enough white around structure) and copy to OpenAI's Sora, to convert it to icon style. Use the prompt "reduce complexity in the details, paths, and outline in this shape, but keep the overall shape in the same form"

1100

My pipeline is like this: go to ChimeraX, load your protein and change the following display settings

1110