Skip to Content

Biological News

Antibody Design, Publicly Challenged

Comes now some rather disturbing news in the antibody field. These things are extremely important, both as therapeutics and as research reagents, and developing them for either purpose is no stroll down the garden walk. There are a number of techniques for raising and producing antibodies (see that first link), but they all have their complications and limitations. The two biggest branches of that tree are to expose humans (or other animals, including mice that have had their immune systems humanize) to some antigen or infectious agent and isolate the relevant B cells, or to go in vitro and express as much IgG diversity as you can via something like phage or yeast display. That’s a grievous oversimplification on my part, as a brief look at just monoclonal antibody production will show you, but that’s a start.

As the paper under discussion today mentions, though, there’s another very different possibility: computing your way to the antibody you need. That is, to put it lightly, nontrivial – we know more than we ever have about antibody structure and function, but knowing enough to design them from scratch is something else. Still, there have been reports. Two in particular stand out, from Ram Sasisekharan and co-workers at MIT. The second of those papers especially talks about “rational engineering” of an anti-Zika antibody via computational epitope selection, and since rational engineering has not exactly been the key step in antibody work historically, it stands out.

Complications have now ensued. The two papers just referenced do not, interestingly enough, provide a sequence for the new antibodies they discuss. The authors of the new critical look at their work, with affiliations at Adimab, Dartmouth, and MIT, point out that this is an odd thing for any journal to permit, since the idea is that others will be able to reproduce reported work, right? At any rate, they’ve cross-referenced patent applications and GenBank sequences to the published papers from the Sasisekharan group and have arrived at what they believe is the sequence of the computationally derived antibodies. And you know what? They’re extremely damned similar to antibodies that had been reported before, in each case:

We present with a high degree of confidence the actual sequence identity of the designed antibodies, and a more plausible genesis narrative. Comparisons of these sequences to those of previously described human B cell-derived antibodies to the same targets show striking similarities. By contrast, those designed sequences appear very dissimilar from the templates said to have been used to start the design process (Figure 2 and S4); we leave it to the reader to judge the likelihood of these highly homologous sequences being re-discovered coincidentally, or simply derived from existing antibodies targeting the same epitopes as those of the computationally designed antibodies.

For example, the supposedly engineered anti-Zika antibody contains an oddly unnatural bit that turns out to have also been in the previously reported antibody. It was, in that older case, an artifact of the expression vector and is not related to epitope recognition at all. What, the new paper asks, are the odds that you would compute your way to that? (While not deliberately introduced as such, this reminds one of the use of trap streets in the mapmaking business). The paper finishes up with what are without doubt fighting words: “We find it difficult to view these authors’ approach in any light other than an intent to mislead as to the level of originality and significance of the published work.”

The lead author of the new paper, Tillman Gerngross, is not mincing words when asked for comment, either. Here he is talking to Stat:

“We looked at exactly two cases, and in both did we find irregularities. To me, if you’re sitting in the kitchen and two fat cockroaches walk across the floor, what’s the chance that there’s only two?”

Yeah, this is glove-across-the-face material, and no mistake. And there are business implications as well. Visterra is a company focused on antibody engineering, and it was founded out of the Sasisekheran lab. Otsuka acquired them recently for $430 million, and this new report can’t make anyone there very happy. There will be more on this story – pretty sure of that – and it doesn’t look like one of those disputes that will be resolved quietly. Not when it starts off like this.



29 comments on “Antibody Design, Publicly Challenged”

  1. Former MITer says:

    Dane Wittrup, one of the co-authors of the MABS paper, and Ram Sasikeharan are both intramural faculty at the Koch Institute of Integrative Cancer Research at MIT and also tenured faculty of Biological Engineering at MIT. Their respective labs and offices are on the second and fourth floor of MIT’s Building 76 (Koch Institute). Would be interesting to be present at faculty meetings.

  2. MassMab says:

    Grab the popcorn and sit back. One of the authors of the accusatory paper is no stranger to hyperbole and disingenuous behavior. Ego and $$$$.

  3. Passerby says:

    “To me, if you’re sitting in the kitchen and two fat cockroaches walk across the floor, what’s the chance that there’s only two?”

    Not sure that’s the right analogy. A Bayesian approach to the problem would have to take into account the prior distribution of fat cockroaches, and that could impact the chances quite a bit. Much will also depend on factors other than body weight.

  4. Frank says:

    This adds another layer of doubt on the AI solving everything in pharma hype. How can the in silico paper be published without showing the antibody sequence somewhere?

    1. M says:

      The key paper discussed has a strong computation element, but on my skimming of it isn’t well described as “AI” (that’s not the computational work) or in silico (there’s empirical data).

  5. Anyway, the hard part is not in designing antibodies, but in predicting whether they will work on patients in the clinic. For this task, computers can help a bit, and I wrote about it in my blog today (for the case of immuno-oncology antibodies):

  6. DriveBy says:

    Looks like we have a front-runner candidate for the 2019 Sirtris Medal (won’t be the first or last time BS research emerges from the Boston area…

    1. Isidore says:

      Oh, I don’t think the Boston/Cambridge area generates any more BS research than other areas proportionally to the amount of overall research being carried out, it’s just that there is an awful lot of research going on in universities, medical schools and biopharma companies in the area.

      1. DriveBy says:

        It’s the impact factor, not the proportional amount – the reputational clout of the academic centers coupled with the synergy of big VC money leads to outsize effect. The biotech version of “Silly Valley” from the dotcom era, which is the best analogy for biotech today. That’s harder for other places (e.g. the midwest) to pull off.

  7. Chrispy says:

    It is quite common for papers with antibodies to not include their sequences. There was a time when people were using polyclonals or uncharacterized hybridomas, but this time has passed. Now it is standard for antibodies to be cloned out and sequenced and then generated recombinantly — this is akin to total synthesis “proving” a structure by generating it synthetically. Editors (and readers) need to insist that antibodies have characterization that includes full sequences, and antibodies from hybridomas need to be made recombinantly to demonstrate that the sequence is correct.

    1. TruthOrTruth says:


      It is unacceptable, yet common, to find papers that publish VDJ gene usage and CDRH3 sequences only. The full sequence should be a requirement for publication.

    2. Ted says:

      Indeed. The sequences are basically never available when working with commercial antibodies, either. As an organic chemist that frequently makes derivitized mAbs for diagnostics, I find this endlessly frustrating. MS services are on the order of 4 – 8K and chew up 0.5mg, so it’s tough to justify when you might be screening a dozen candidates. I’ve always chalked it up to biologists willingness to work with more uncertainty than chemists…


  8. Anonymous says:

    As an aside: Wow! Gerald Wogan (MIT, emeritus) is on the Sasisekharan (MIT) paper. Wogan (tox, then part of “Food and Nuts” in building 56) was a collaborator of George Buchi (chem, building 18) on things like the aflatoxins going back to the 1960s.
    That’s all. Back to the controversy.

    1. Isidore says:

      The corresponding author of the MABS “Perspective” also spent time at MIT, in a couple of different labs, as a post-doc and visiting scientist (as per Wikipedia) and another of the authors, as noted above, is a senior faculty member of MIT. This has all the markings of an intramural argument and like civil wars these tend to be quite nasty.

  9. eub says:

    “we leave it to the reader to judge the likelihood”

    Oh that’s never a sign of peace and harmony.

    Those in the field, is there a plausible way this sequence homology could have come out of the process described in the previous paper? The work was in silico so there’s no “oops, our PCR picked up a contaminant carried on my bowtie”, right?

    1. A Knowing Mess says:

      That’s exactly the problem – the parental antibodies were used as a starting point to design new antibodies against the target. However, if you start out with an anti-HA antibody, make some changes, and wind up with an anti-HA antibody, how much novelty are you actually introducing in the design process? That’s the criticism – most of the AA sequence diversity from the ‘new’ antibodies to the parental (?) antibodies comes in the framework regions and not in the CDRs.

      1. eub says:

        Ah, thank you for the context. That makes more sense in personal terms than widespread outright cheating, but yeah, what you said.

      2. Carl_Bar says:

        Thank you for the explanation the blog post was a bit technichial in places and i was having trouble deciphering exactly what they where supposed to have done.

  10. Anonymous says:

    Derek, I wonder if you plan to post of the paper making headlines (including the NY Times), claiming that indole-3-carbinol is a WWP1 inhibitor, thereby reactivating the tumor suppresor PTEN ( Check out Figure 5B. It is, ahem…”interesting”.

    1. FitAnyCurveifyoutryhardenough says:

      I would love to see how hard they tried to fit that curve. LOLZ

  11. xtalographer says:

    Dane Wittrup is a phenomenal and extremely careful scientist. No way he makes this type of accusation lightly.

    1. Joe2 says:

      Couldn’t agree more!!!

  12. Ted says:

    I have to admit, I’m a little mystified as to why you would stick with antibody structures when rationally designing a protein binding element anyways. People spend a lot of time trying to get away from the baggage that comes with IgGs…


  13. Greg Czyryca says:

    Present day protein design usually involves knowledge-based protocols (as opposed to true de novo design, based on more fundamental principles). It is therefore not surprising that the “designer” proteins contain known fragments.

    Also, the more machine learning is employed, the more prior knowlege bias will be in the results. True de novo design, conversely, employs evolutionary strategies instead, and cannot IMO benefit much from AI.

    1. Skeptic says:

      It’s a pity that the term “AI”, which used to mean rule-based algorithms, now usually means black-box neural networks.

  14. DiGRAS GRASoxide says:

    Leaders of the scientific community making public their private little petty feuds…..the guy who made the obscene stupid analogy equating data points to cockroaches should resign

    1. Skeptic says:

      I agree that the cockroach remark was needlessly inflammatory, but I think scientific debates should be conducted out in the open.

      I don’t know why you think an allegation of scientific fraud is “petty”.

      1. Scott says:

        Very much agree that an allegation of outright fraud, *that the method described in the paper is not what gave them their end result(!)* is anything but ‘petty’.

        I’d go so far as to say it’s rather the opposite of petty. Gross.

  15. dipthroat says:

    As I like to say
    cheating is the most effective way to make career in science.
    Or, any other field really

Leave a Reply

Your email address will not be published. Required fields are marked *

Time limit is exhausted. Please reload CAPTCHA.