Business
We evoke media theorist Steven Johnson’s
provocative, engaging and surprising examples of feedback, self-organisation and adaptive learning in influencing the evolution of ‘emerging systems’. The power of self-organisation ushers in a revolu- tion every bit as significant as the introduction of electricity. In explaining why, the whole is some- times smarter than the sum of its parts, Steven Johnson places ‘self-organisation’ on the front lines of this exciting upheaval in science and thought1. Let us think about this further. Every once in a
while, within a complex technical challenge, some experiment or observation may yield a surprising, transformational result. It may be good news, per- haps in the form of a known drug-like entity show- ing promise in treating a challenging phenotype, or the finding could be a grave disappointment, such as problematic toxic side-effects. Whenever the unexpected happens, our instinc-
tive response is to seek clarity. What caused the surprise? This can certainly lead to new experi- ments, but wise researchers typically turn first to prior data and careful review of existing literature. In many cases, this informed re-examination of prior records will suggest a very plausible and rational explanation. One might then ask, if information was already
in place to rationalise the surprise, why did we not foresee it? Perhaps hindsight is 20:20. Human brains are
better equipped to interpolate a known observa- tion back to a set of prior contributing causes than to extrapolate complex conditions ahead to an eventual consequence. We can easily imagine that striking a glass vase will produce a mess of shards, but if we encounter only the shards without the proper context, it is harder to visualise the original vase that they once formed. By analogy, a mess of non-contextual data rarely shout out ‘imminent drug failure’, but once a drug has failed, the shards of prior evidence can be seen everywhere. That seems self-evident, but is it also artificially
self-limiting? Let us consider the following: 1) All information that once defined the vase remains contained in the mess of shards. Similarly, it is often true that key evidence to suggest a sur- prising biomedical observation was already known, just not put in the right context. 2) There are savants who can sift through a mess of glass shards and posit the original shape and pattern of the originating vase. Analogously, for many surprising research discoveries, there later prove to be people who, quite verifiably, can say: “I told you so.” 3) Finally, while many AI algorithms demonstrate
Drug Discovery World Summer 2018
pattern recognition powers to resemble or exceed human savants, various classes of algorithms such as Deep Learning (DL) and Analysis of Emergent Behavior (AEB), have a capacity for assimilating the relationships between disparate shards and plausibly unifying such information toward impor- tant real-world conclusions. Knowing this, why must immensely capital-
intensive and labour-intensive drug design projects still rely so heavily on luck and surprise? Part of the answer is cultural. Chief Scientific
Officers and programme directors may be comfort- able making decisions based on flawed and incom- plete data that they understand, but are often reti- cent about making comparable decisions based on complex reasoning whose subtleties exceed normal human cognition. Thus, when it comes to advice from savants and arcane algorithms, pharma execs tend to be more skeptical than the average sports gambler. One wonders who is more likely to hit the big payoff? That natural bias, however, is beginning to give
way in the face of success. As has been discussed in a pair of recent editorials4,5, emerging algorithms have demonstrable capacity for assimilating multi- modal/multisource data via associations that almost mimic human intuition. Just as the vase is defined by shards of many sizes and shapes, DL can extract information from diverse sources (data, metadata, annotations, accompanying graphics, text mining, etc). Just as the shattering process may have scattered in directions that barely reflect the original vase, DL recognises that the placement of no individual shard dictates the vase, but rather seeks to find ways to fit those shards together into a harmonious, self-consistent originating behaviour. In practice, DL6 and AEB7 are very useful addi- tions to the biotech arsenal, yet DL in particular has
43
Figure 1 Predictive inference Intuitive: known condition rationalised by many diagnostic observations. Challenging: diagnostic observations predict ‘knowable’ condition
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76 |
Page 77 |
Page 78 |
Page 79 |
Page 80
