不用化疗，一家公司正利用人工智能开发治癌新方法

    我们是否真的在两三年之后，就能实现治愈癌症的愿景？波士顿小型生物科技公司Berg的总裁尼文•纳雷因表示，可能真是这样。

    凭借亿万富翁、房地产业大鳄卡尔•伯格和米奇•格雷提供的资金，纳雷因和他带领的科学家、技术人员和编程人员团队耗时6年，完善并测试了一个人工智能平台，纳雷因认为，这个平台可能很快就会解开癌症的密码，同时为治疗包括帕金森症在内的一系列严重疾病提供有价值的信息。

    凭借着跟多所大学、医院甚至美国国防部建立的合作关系，伯格公司及其超级计算机系统已经分析了成千上万的病历和组织样本，以找到有可能全新的药物靶标和生物标志。

    经过庞大的数据计算，伯格公司开发出第一款新药——BPM 31510，目前该药已经进入临床测试阶段。它可以重组癌细胞的新陈代谢，重新教会癌细胞如何死亡。在这个过程中，癌细胞就会自然死亡，使患者不必经历对身体伤害极大又十分昂贵的化疗过程。

    到目前为止，伯格公司的主要资源都集中在前列腺癌上，因为目前有大量关于前列腺癌的数据可供研究。不过拜一项最新合作所赐，该公司现在已经开始构建针对胰腺癌的新模型了。胰腺癌也是最凶险的癌症之一，目前的存活率只有7%。

    这个目标本身可谓雄心勃勃，但它还只是冰山的一角。除了治疗前列腺癌和胰腺癌之外，伯格公司还希望分析多种其它疾病的数据，包括乳腺癌。另外，伯格公司还认为，它的人工智能平台可以根据病人的特异性制定专门针对个别患者的治疗方案，从而将掀起一场药物测试的革命，并显著降低药物的负面作用在临床实验和医疗实践中的风险。

    我采访了卡尔•伯格和纳雷因，探讨了该公司的工作机制，以及他们在未来几年内的目标。以下是采访摘要。

    财富：卡尔，你为什么选择从房地产业转向医疗行业？它的进展是否符合你的预期？

    卡尔•伯格：我已经在风投界干了40年了，但我从来没有触碰过生物科技领域，因为我担心与政府审批有关的风险。做风投本身就不容易，又要多花一番工夫去获得美国食品药品监督管理局的认证，那就会更难。但大概8年前我曾说过，现在我不必再做一堆小公司了，而是有能力做一些影响力足够大甚至有希望改变世界的事。这个目标激励了我，然后我认识了尼文，我们就是这样开始这项事业的。

    是尼文说服了你进入医疗行业，还是你找到了尼文？

    卡尔•伯格：当时我正考虑投资一款护肤产品，然后我在迈阿密大学经人介绍认识了尼文。尼文当时是那个项目的经理，那个项目开始大约一两个月后，尼文给我打电话说：“卡尔，这款护肤产品似乎对治疗癌症有效。”我说：“好吧，如果你治好了谁，记得让我知道。”

    你听起来好像不太相信。

    卡尔•伯格：人人都知道，每种癌症都是不一样的，那么这个东西怎么会有效呢？在我看来根本就说不通。这时尼文说：“我能飞到加州向你展示一下我的成果吗？”然后他就来了，经过一番交流，我相信他使用的技术和方法真的有可能在医药市场掀起一场革命。

    尼文，你是怎样让卡尔•伯格相信，你那款护肤产品上有可能治愈癌症？

    尼文•纳雷因：当我见到卡尔时，我们原则上同意，肯定有办法建立一个更高效的医疗系统，它能够以非常精确的方式，将病人与正确的药物进行匹配。卡尔支持我们将这个理念引向深入。我们不是利用筛选过的化学制品治疗病人，而是从人体的细胞样本入手去了解人体生物学，然后据此研发药物的。我们使用的是人工智能，而不是各种假设。

    人工智能究竟在这个过程中起了什么样的作用？

    尼文•纳雷因：如果你从一个假设入手，你就排除了很多其他可能产生真正效果的领域。有多少次药物在晚期测试的失败，是因为它的早期科研不够扎实，或是选择了错误的靶标？

    在伯格公司，我们只针对一个组织样本就建立了超过14万亿个数据点。无论是使用人力，还是使用传统的推理假设模型，要想从所有这些数据中摘取有价值的信息，都是不可能的。所以当我们构建我们所称的疑问型生物平台时，我们使用了人工智能来分析所有数据。人工智能可以从病人的生物数据、临床样本和人口统计资料中摘取所有的信息，并且可以根据类似性和差异性进行分类和分层，从而帮助我们了解健康细胞和病变细胞之间的差异。

    14万亿个数据点听起来有点超负荷的感觉。

    尼文•纳雷因：所以它有两个组成部分：首先是生物信息，然后还有所谓的“组学”。我们不仅仅是分析基因组，而是研究一个组织样本的所有基因、蛋白质、代谢分子、脂质、病历记录、人口统计学资料、年龄、性别等等信息。我们把人体的3万个基因与6万种蛋白蛋和几千种脂质、代谢分子的信息综合起来，然后把这些成分用具有机器学习功能的高阶数学算法进行计算，以了解它们的各种关联性和相关性。

    组学是一个相对较新的术语，它意味着你不能仅仅盯着基因组，而是所有的“组”——比如蛋白质组、代谢组等等。虽然可能我们出生就带着3万个基因，而且这些基因可能还有某些天生的突变，但这并不是故事的结尾。你住在纽约市，暴露在环境中的不同物质里，你的饮食与阿拉巴马州的某个人不一样，你的睡眠习惯也与犹他州的某个人不一样。所以我们认为，这些东西应该综合起来，才能完整描绘你的“组学”，即你的整体资料。

    但是这些东西怎样让我们治病？看起来只是一堆数据分析而已。

    尼文•纳雷因：我知道你经常报道航空业，你可能很熟悉航空公司的路线图了，它们展示了各个枢纽城市和目的地之间的联系。在我们的疑问型生物平台上，所有这些数据分析的结果看起来就像3D版的航空路线图。但这些联系并不是城市与城市之间的，而是基因与蛋白质之间。然后我们把重点放在那些大的枢纽上，看看是否出了什么问题。比如如果达拉斯市是在俄克拉荷马州境内，我们都知道肯定有问题，这时人工智能就会把达拉斯推回北德克萨斯州，然后分析生物学中的哪些事件可以让人体重启正常的流程。这就是我们的研究重点，即生物的基本元素，以及能让健康流程重启的基因和蛋白质。

    在真实世界中，你利用该平台取得过成功吗？

    尼文•纳雷因：我们正在测试一款名叫BPM 31510的药物，它就是我们利用疑问型平台研发的。目前显示的结果非常令人鼓舞。该平台显示，新陈代谢越多，治疗就会越有效。根据我们对患有某些癌症的病人的观察，的确是这样。比如我们在一名患有膀胱癌的病人身上测试了这款药物，膀胱癌是一种非常凶险的癌症，几乎对所有疗法都没有反应。我们在他身上使用了BPM 31510，该药以癌细胞的新陈代谢为靶向，到了第18周，他的肿瘤已经完全消失了。

    这种疗法取得专利了吗？

    尼文•纳雷因：我们把前六年的大部分时间花在构建平台、研究各个重点领域、对早期药物进行临床实验和实现技术使用的多样化上。我们在全球已经注册了500多个专利。所以我们在生物学、数学、信息学上都有专利，对每个个体生物指标和药物靶标也都有专利。总之我们有着非常坚实的知识产权资产。

    你们的竞争对手是谁？与他们相比，你们在今后的发展中处于何种地位？

    尼文•纳雷因：我们经常会被问到这个问题。也有一些人和机构在做我们正在做的事。他们是一些蛋白质和分析学上的顶尖公司，但我们目前还没有发现哪家公司把有关的生物学、组学研究和临床能力整合到一个疑问型平台上，来对人体产生坚实的理解，并以一种新的方式开发药物。另外，我们是用数据产生假设，而不是用假设产生数据，所以它是一种不同的方法。我们在这方面还是挺独特的——无论是在技术上还是商业上。

    卡尔，过去几年里，你和米奇•格雷一直是伯格公司的唯一投资人，为什么会这样？

    卡尔•伯格：如果你在这些东西的早期阶段就让太多人进入，尤其是这个项目又有比较高的风险，那么你基本上肯定会失败，因为只要有什么事情出了差错，人们就会感到沮丧和担心。凭借多年的风投经历，我基本上已经处变不惊了。就算出了大乱子，我也不会那么沮丧。我知道那就是你需要预料到的。

    你们现在打算开放融资了吗？

    卡尔•伯格：我们当然希望做些其他事情，并且引入新的投资人。但我们希望在此之前先达到某一个点。我认为我们离那个点已经非常近了。（财富中文网）

    译者：朴成奎

    审校：任文科

    Could we be just two or three years away from curing cancer? Niven Narain, the president of Berg, a small Boston-based biotech firm, says that may very well be the case.

    With funding from billionaire real-estate tycoon Carl Berg as well as from Mitch Gray, Narain, a medical doctor by training, and his small army of scientists, technicians, and programmers, have spent the last six years perfecting and testing an artificial intelligence platform that he believes could soon crack the cancer code, in addition to discovering valuable information about a variety of other terrible diseases, including Parkinson’s.

    Thanks to partnerships formed with universities, hospitals, and even the U.S. Department of Defense, Berg and its supercomputers have been able to analyze thousands of patient records and tissue samples to find possible new drug targets and biomarkers.

    All this data crunching has led to the development of Berg’s first drug, BPM 31510, which is in clinical trials. The drug acts by essentially reprogramming the metabolism of cancer cells, re-teaching them to undergo apoptosis, or cell death. In doing so, the cancer cells die off naturally, without the need for harmful and expensive chemotherapy.

    So far, Berg has concentrated most of its resources on prostate cancer, given the large amount of data available on the disease. But thanks to recently announced partnerships, the firm is now building a new modeltargeting pancreatic cancer, which is one of the deadliest forms of cancers with a survivorship rate of only 7%.

    Ambitious as that may be, it is really just the tip of the iceberg. In addition to mapping out prostate and pancreatic cancer, Berg hopes to analyze data from a whole host of other diseases, including breast cancer. Additionally, Berg thinks his company’s artificial intelligence platform can also revolutionize drug testing by creating individualized patient-specific treatment options, which he believes will ultimately reduce the risk of adverse drug interactions in clinical trials and hospitals by a significant degree.

    I sat down with Berg and Narain to discuss how the company works and what they hope to accomplish in the next few years. The following interview has been edited for publication.

    Fortune: Carl, why did you decide to move from real estate into healthcare and has it panned out like you thought it would?

    Carl Berg: I have been in the venture capital business for 40 years but I never once touched biotech because I was concerned about the risk associated with government approval – it’s bad enough when you’re doing venture capital but adding one more equation, like getting approval from the FDA [Food and Drug Administration] makes it a lot harder. But about eight years ago I said, instead of getting into a whole bunch of small companies, I am in a position now where I can do something really big in a hope that it changes the world. So that’s what motivated me, and then I met with Niven, and that’s what got it started.

    Did Niven convince you to go into biotech or did you find Niven?

    CB: I was considering a skin care product investment and I was introduced to Niven at the University of Miami. Niven was the project manager and about a couple months into work on this product, Niven called me and said “Carl, this skin care product appears to have an effect on cancer.” To which I said “Sure, whenever you cure somebody, let me know.”

    You didn’t sound very convinced.

    CB: Everybody knows that every cancer is different, so how could this one thing work? That didn’t make any sense to me. And Niven said, “Can I fly out to California and show you my results?” And he came out, and we talked, and I got convinced that the technology he was using and the approach he was taking, could revolutionize the pharmaceutical market.

    Niven, what did you say to convince Carl Berg that your work on skin cream could possibly lead to a cure for cancer?

    Niven Narain: When I met with Carl we were aligned philosophically that there has to be a better way to create a more efficient healthcare system – one that really matches the right patients to the right drugs in a very precise manner. So Carl supported taking this concept to the next level. Instead of treating humans with chemicals, that are screened to become drugs, we actually started with human tissue samples and work to understand the biology and develop drugs based on that. Using AI [artificial intelligence] instead of hypotheses.

    How exactly does artificial intelligence come into play here?

    NN: When you start with a hypothesis, you are dismissing a lot of other areas that might actually have an impact on whatever you are trying to figure out. How many times do we see drugs get to late stage trials and fail because the early science either wasn’t robust enough or focused on the wrong target?

    At Berg, we use AI to create over 14 trillion data points on only one tissue sample. It is actually humanly impossible to go through all this data and use the traditional hypothesis inference model to glean any value out of all of it. So early on when we built what we call an interrogative biology platform using AI to go through all that data. AI is actually able to take all the information from the patient’s biology, clinical samples, and demographics and really categorize which ones are similar and which ones are different and then stratify those in a way that helps us understand the difference between the healthy and diseased.

    Fourteen trillion data points sounds like information overload.

    NN: So there are two components: the upfront biological and there is something called omics. We go much deeper than just analyzing the genome, we look at all the genes in that tissue sample, all the proteins, metabolites, lipids, patients records, demographics, age, sex, gender, etc. We combine the 30,000 genes in the body with about 60,000 proteins and a few hundred lipids, metabolites. Then we take those components and subject them to high order mathematic algorithm that essentially learns, uses machine learning, to learn the various associations and correlations.

    Omics – it’s a fairly new term. It means you’re going beyond just the genome. It means all the omics – proteomics, metabolomics, and proteins. So we may be born with 30,000 genes, and those genes were born with certain mutations, but that’s not the end of the story. You live in New York City, you are exposed to different things in the environment, your diet is different than someone who lives in Alabama and your sleeping habits are different from some who lives in Utah. We believe all of these things have to be put together to tell the whole story of your omics – the full profile of you.

    But how does all of this get us to a cure for anything? Seems like a bunch of number crunching.

    NN: I know you cover the airline industry pretty intently, so you are probably familiar with those airline route maps that show all the connections between hubs cities and destinations. So with the interrogative biology platform, the result of all that number crunching looks similar to a 3D version of those maps. But instead of those connections going between cities, they are going between genes and proteins. We then focus in on the big hubs and see what, if anything, is wrong. For example, in a system, if Dallas is in Oklahoma, obviously we know something is wrong, so the AI helps to push Dallas back into North Texas, and analyze what events happened in the biology to make that a normal process again. This is what we focus in on. The elements within the biology, the genes and proteins that made that a healthy process again.

    Have you had any success using this platform in a real world situation?

    NN: We are in clinical trials for a drug, BPM 31510, which we developed using the interrogative platform. The results we have seen so far have been very encouraging. The platform predicted that the more metabolic, the better the treatment will work. And that is exactly what we are seeing in patients for certain types of cancer. For example, we tested this on a patient who had bladder cancer. It was a very aggressive cancer, which failed to respond to all other therapies. We then put him on BPM 31510, which targeted the metabolism of the cancer cell, and by week 18, the tumor was completely gone.

    Is this a patented process?

    NN: We spent the lion’s share of the first six years building the platform, developing it into various areas of focus, getting our early drugs into clinical trials and diversifying the use of the technology. And we have filed over 500 patents around the world that govern this specific elevated biology. So we have patents on the biological process, on the mathematics, the informatics, on each individual candidate biomarker, and drug targets. It is a very robust IP portfolio.

    Who are your competitors and where are you versus them in taking this process to the next level?

    NN: We get asked that fairly often. There are folks and entities that do pieces of what Berg does. They’re leading companies focused on proteins or analytics, but there isn’t one company we can identify or know of that has taken the biology, the omics, the clinical capability and put it all into an interrogative platform to really allow for a robust understanding of the biology to discover drugs in a different way. Also, we are allowing the data to generate hypotheses instead of hypotheses generating data, so it’s a really different approach. We are fairly unique in that respect – both from a technology, but also from a commercial standpoint.

    Carl, for the last few years, you and Mitch Gray have been the only investors in Berg. How come?

    CB: I’ve learned that if you get too many people in the early stages of these things, especially within something as risky as this was, basically you have failed because people get upset and they get worried when anything goes wrong. Through all the years that I have been doing this I can kind of roll with the punches. If something goes haywire it doesn’t upset me that much. I know that’s what you’re going to expect.

    Are you ready to open things up now?

    CB: We are definitely planning on doing some other things and bringing in other investors, but we thought we ought to get to a certain point before we did that. I think we are now very close to that point.