Skip to Content

Cancer

Artificial Intelligence, You Say?

Here’s a story from the Telegraph about a small company (Berg Pharmaceuticals), whose headline certainly got my attention: “Cancer drug development time halved thanks to artificial intelligence”. That set off some alarm bells for me, and not because I fear being replaced by a bearded AI quoting mispronounced Latin tags and John Cleese dialogue. (That would be unnerving). No, it’s that I think that (at this point, anyway) that any group claiming great advances due to artificial intelligence is probably hyping their results. So what’s Berg up to?

Their drug is BPM31510, and you can read through the whole newspaper article without encountering the phrase “Warburg effect“. That’s the target, though, and it’s been a popular one in recent years. There are indeed many tumors that have abnormally high rates of glycolysis (followed by lactic acid fermentation, as opposed to the usual pyruvate pathway). There are a lot of compounds that have been looked at (and are being looked at still) to interrupt this process (the famous dichloroacetic acid is one), and metabolic targeting is considered to be a good-sized field all its own in oncology.

Berg is trying to normalize mitochondrial function, in the hopes that this will derail tumor cell metabolism. It should be noted, though, that there’s still an active debate about whether mitochondrial dysfunction causes the high glycolysis rates, or whether those cause mitochondrial dysfunction. By targeting the mitochondrial end of things, rather than the glycolysis itself, you taken a side in that argument whether you realize it or not.  So how are they targeting them? With ubidecarenone / ubiquinone, also known as Coenzyme Q10. This, of course, is a naturally occurring compound in the body, and is available at any health food store. It is certainly very important to mitochondrial function, and supplementation with it would seem an obvious thing to try if you’re trying to restore that activity. Such supplementation has been tried across a wide range of conditions, but it’s worth noting that ubiquinone has been studied many times as a potential therapy or adjunct therapy in cancer.

So when a newspaper article refers to a new drug being developed, it might be worth mentioned that it’s actually quite an old drug, an endogenous substance very much like a vitamin, whose safety profile is not in doubt, and which has been studied before in the same therapeutic area. Berg does have a new liposomal formulation of ubiquinone, which may well be helpful (and which does require its own safety evaluation, although you’d go into this expected success). Where’s the artificial intelligence come in?

Berg Health’s team used a specialised form of artificial intelligence to compare samples taken from patients with the most aggressive strains of cancer, including pancreatic, bladder and brain, with those from non-cancerous individuals. The technology highlighted disparities between the corresponding biological profiles, selecting those it predicted would respond best to the drug.

“We’re looking at 14 trillion data points in a single tissue sample. We can’t humanly process that,” said Niven Narain, a clinical oncologist and Berg co-founder. “Because we’re taking this data-driven approach we need a supercomputer capability.”

It’s been a while since I heard a biopharma company talk about having to use a supercomputer, I have to say, and I wasn’t sure that the term was still around. The company is looked at proteomic, metabolomic, lipidomic (etc.) profiles of cells and tissue samples, and trying to build a model of the disease state around them. This Bio-IT World article is the best description I’ve found, and the best look at the company in general (the word “supercomputer” does not appear). They’re specifically targeting endogenous molecules – no screening, no synthesis (and thus, as that article notes, no composition-of-matter patents, either). They are mechanism-agnostic, at least at first, which might explain why the Telegraph article is all about BPM31510 restoring mitrochondrial function, while this recent AACR abstract from the company is about its effects on the fluidity of the cell membrane as a mode of action.

To be sure, no one will care so much about the mechanism if the stuff works. I can see two studies on BPM31510 at clinicaltrials.gov at the moment, both of them Phase I. The company was previously working on a topical formulation for squamous cell carcinoma, but their Phase II on that one completed five years ago, so it presumably isn’t going anywhere. Their approach is ambitious, although certainly not crazy, but they’re going to want to keep an eye on their press coverage. Talking up “artificial intelligence” and “cutting development times in half” may bring them a bit more notoriety than is needed.

10 comments on “Artificial Intelligence, You Say?”

  1. Peter Kenny says:

    At least the ToryGraph has a bit more credibility than the Daily Fail?

  2. a says:

    It is clearly an advertorial. Not even worth reading.

  3. Isidore says:

    “The company is looked at proteomic, metabolomic, lipidomic (etc.) profiles of cells and tissue samples […]”
    Having been involved very actively in one of the above areas for longer than its name, in academia and in the biopharmaceutical industry, I can say unequivocally that many, many academic and industry research groups are doing exactly that. In fact they also are doing genomics, transcriptomics, glycomics, and any other “omics” that may be of even the slightest relevance to the research. To my knowledge nobody (other than Berg, apparently) is using a “supercomputer” or “artificial intelligence” to examine, reduce and correlate the various streams of data.

  4. Ash (Curious Wavefunction) says:

    I don’t think Berg Pharmaceuticals is doing anything novel. Having said that, they don’t seem to be doing something obviously stupid either. As the commenter above pointed out, many academic and industrial groups are trying to use machine learning/”AI” to identify pharmaceutically relevant genetic differences between cancerous and non-cancerous tissues. The issues as always lie in the well-trodden and yet ill-understood domains of false positives, the correlation vs causation fallacy and trying to unearth a small amount of causal signal in a river of noise. It won’t be easy and its shouldn’t be hyped as a cure-all. That being said, I’m not complaining that more people rather than fewer are working in this area and trying to separate the wheat from the chaff.

  5. marcos says:

    Did you see who is on the Berg council of advisors? Good old Ben Carson!

  6. Kent G. Budge says:

    “an endogenous substance very much like a vitamin, whose safety profile is not in doubt”

    Given the rotten bioavailability of Q10, particularly by the oral route, it would have a hard time hurting you even if it was determined to.

    Which is a shame. I’d love to think the Q10 I take is anything more than a charm to keep statin side effects away. And probably as effective as most charms.

  7. MolecularGeek says:

    I’m with Ash on this one. They’re not doing anything that a lot of other startups aren’t also doing, but they aren’t doing anything stupid with it, either. You don’t see much about supercomputers these days, because they were essentially supplanted 15 years ago by distributed high-end commodity server clusters. There are a few fields where putting lots of processors in a single case with shared memory and a printed circuit interconnect that doesn’t need to translate protocols makes sense. But not many. Most companies doing things on a huge scale end up pushing at least part of it into cloud computing where they only pay for the cycles they use instead of the cost of keeping their own internal clusters operational for peak capacity even at slack times. Call it what you want, it’s still mostly about splitting up the problem into smaller chunks so that a lot of computers can work on different portions of it at the same time.

  8. Kumar says:

    Basically Berg is doing lot more bioinformatics using super computer and calling it AI. A savvy advertisement indeed. The real questions are, isn’t there any studies on the molecular mechanism of how Q10 might normalize mitochondrial function, and how effective this approach is, at least at the in vitro level?

  9. Paul Brookes says:

    Along similar lines, Edison (a small pharma looking at mitochondria) used obfuscatory language in their early PR, to describe their drug EPI743, which turned out to be a simple tocopherol analog. The grating phrase for mito’ biologists was “redox encrypted molecules”. As if redox potential is some obscure property that can’t be measured.

    Another example is Novelos, with their “NOV-002” drug, which turned out to be nothing more than glutathione disulfide (GSSG). Trials went nowhere, Cellectar bought the drug and is doing more trials on it.

    At the extreme end of things is Bruce Ames with Juvenon, a proprietary mix of acetyl-carnitine, alpha lipoic acid and biotin, plus a dose of resveratrol, green tea extract and quercetin for good measure. Yours for only $40 a bottle.

    On the straight-talking side of things, there’s Antipodean pharma with TPP-conjugated “Mito-Q”, which at least has a decent rationale and mechanism of action. However, several clinical trials went nowhere and Mito-Q is now an ingredient in skin cream.

    And so to Berg… aside from the AI speak, at issue are the hundreds of clinical trials on Q10 and related redox active molecules (lipoic acid, glutathione, idebenone, etc) for the enhancement of mitochondrial function in diseases where it’s actually depressed – i.e. the inherited mtDNA mutations. Most mito’ patients take Q10 but it’s little more than a good luck charm as others have mentioned. There isn’t a whole lot of rationale for how Q10 would doing anything differently (i.e. something other than not much at all) in a cancer cell. If the drug can’t improve function in bona fide cases of mito’ dysfunction, why would it do so in cases where mito’ dysfunction is maybe not even there to begin with?

    Then of course, given that many existing anti-cancer agents specifically work by oxidizing things, alkylating DNA and other targets, one has to question the wisdom of throwing into the mix a bunch of small molecules that do the opposite. If your redox active compound blunts the therapeutic potential of a go-to therapy such as cisplatin, it’s unlikely to be popular with cancer physicians.

  10. Patrick Sweetman says:

    Genetic algorithms, as used by Eureqa have been called AI and have been used to concoct systems models from genomics data.

    The software at http://www.nutonian.com/products/eureqa/ used to be free but has been developed a bit since those days and is only free now on a trial basis.

    It is a brilliant and amazing tool/toy to play with.

    No interests to declare.

Comments are closed.