TC, 0000-0002-5577-3516; DSH, 0000-0002-3012-7446;
BKB, 0000-0002-6700-1468; AAK, 0000-0003-4563-3226;
BTD, 0000-0003-4992-2623; GPW, 0000-0002-0503-9348;
EF, 0000-0002-8362-100X; P-MA, 0000-0003-1126-1479;
MZ, 0000-0003-0539-630X; MMH, 0000-0002-4517-1562;
WX, 0000-0002-1871-6846; GLR, 0000-0003-1763-5750;
BJL, 0000-0001-8690-9554; JI, 0000-0003-1633-5780; JL, 0000-0003-0811-0944;
SW, 0000-0003-0568-298X; AEC, 0000-0003-1555-8261;
AS, 0000-0002-6443-4671; JX, 0000-0001-7111-4839;
EMC, 0000-0003-3877-0433; CAL, 0000-0002-7762-1089;
SCT, 0000-0003-3247-6487; AMA, 0000-0001-8655-8109;
ZL, 0000-0001-9998-916X; DJH, 0000-0003-3332-9307;
DD, 0000-0001-8931-9461; YQ, 0000-0002-5796-7453; AK, 0000-0003-3084-2287;
YP, 0000-0001-9309-8331; LKW, 0000-0001-6681-9754;
MHSS, 0000-0001-8008-0546; SMB, 0000-0002-1400-3398;
SJS, 0000-0003-2191-0778; AH, 0000-0003-1349-4030;
AG, 0000-0002-5324-9833; CSG, 0000-0001-8713-9213
Deep learning describes a class of ma chine learning
algorithms that are capable of combining raw inputs into
layers of intermediate features. These algorithms have
recently shown impressive results across a variety of domains.
Biology and medicine are data-rich disciplines, but the
data are complex and often ill-understood. Hence, deep learn-
ingtechniquesmaybeparticularlywellsuitedtosolve
problems of these fields. W e examine applications of deep
learning to a variety of biomedical problems—patient classi-
fication, fundamental biological processes and treatment of
patients—and discuss whether deep learning will be able to
transform these tasks or if the biomedical sphere poses
unique challenges. Following from an extensive litera t ur e
revie w , we find that deep learning has yet to rev olutionize
biomedicine or definitively resolv e any of the most pressing
challenges in the field, but promising advances have been
made on the prior state of the art. Ev en though impro v ements
over previous baselines hav e been modest in gener al, the
recent progress indicates that deep learning methods will
pro vide valuable means for speeding up or aiding human
inves t iga tion. Though progress has been made linking a
specific neural network’s prediction to input featur es, under-
standing how users should interpret these models to make
testable hypotheses about the sys tem under study remains
an open challenge. Furthermore, the limited amount
of labelled da ta for tr aining pr esents problems in some
domains, as do legal and priva cy constr aints on work with
sensitive health records. Nonetheless, we foresee deep learn-
ing enabling changes at both bench and bedside with the
potential to transform sev er al areas of biology and medicine.
1. Introduction to deep learning
Biology and medicine are rapidly becoming data-intensive.
A recent comparison of genomics with social media, online
videos and other data-intensive disciplines suggests that geno-
mics alone will equal or surpass other fields in data generation
and analysis within the next decade [1]. The volume and com-
plexity of these data present new opportunities, but also pose
new challenges. Automated algorithms that extract meaningful
patterns could lead to actionable knowledge and change how
we develop treatments, categorize patients or study diseases,
all within privacy-critical environments.
The term deep learning has come to refer to a collection of
new techniques that, together, have demonstrated break-
through gains over existing best-in-class machine learning
algorithms across several fields. For example, over the past 5
years, these methods have revolutionized image classification
and speech recognition due to their flexibility and high accuracy
[2]. More recently, deep learning algorithms have shown
promise in fields as diverse as high-energy physics [3], compu-
tational chemistry [4], dermatology [5] and translation among
written languages [6]. Across fields, ‘off-the-shelf’ implemen-
tations of these algorithms have produced comparable or
higher accuracy than previous best-in-class methods that
required years of extensive customization, and specialized
implementations are now being used at industrial scales.
Deep learning approaches grew from research on artificial
neurons, which were first proposed in 1943 [7] as a model for
how the neurons in a biological brain process information.
The history of artificial neural networks—referred to as
‘neural networks’ throughout this article—is interesting in its
own right [8]. In neural networks, inputs are fed into the
input layer, which feeds into one or more hidden layers,
which eventually link to an output layer. A layer consists of a
set of nodes, sometimes called ‘features’ or ‘units’, which are
connected via edges to the immediately earlier and the
immediately deeper layers. In some special neural network
architectures, nodes can connect to themselves with a delay.
The nodes of the input layer generally consist of the variables
being measured in the dataset of interest—for example, each
node could represent the intensity value of a specific pixel in
an image or the expression level of a gene in a specific tran-
scriptomic experiment. The neural networks used for deep
learning have multiple hidden layers. Each layer essentially
performs feature construction for the layers before it. The train-
ing process used often allows layers deeper in the network to
contribute to the refinement of earlier layers. For this reason,
these algorithms can automatically engineer features that are
suitable for many tasks and customize those features for one
or more specific tasks.
Deep learning does many of the same things as more fam-
iliar machine learning approaches. In particular, deep learning
approaches can be used both in supervised applications—where
the goal is to accurately predict one or more labels or outcomes
associated with each data point—in the place of regression
approaches, as well as in unsupervised, or ‘exploratory’ appli-
cations—where the goal is to summarize, explain or identify
interesting patterns in a dataset—as a form of clustering.
Deep learning methods may, in fact, combine both of these
steps. When sufficient data are available and labelled, these
methods construct features tuned to a specific problem and
combine those features into a predictor. In fact, if the dataset
is ‘labelled’ with binary classes, a simple neural network
with no hidden layers and no cycles between units is equival-
ent to logistic regression if the output layer is a sigmoid
(logistic) function of the input layer. Similarly, for continuous
outcomes, linear regression can be seen as a single-layer
neural network. Thus, in some ways, supervised deep learning
approaches can be seen as an extension of regression models
that allow for greater flexibility and are especially well suited
for modelling nonlinear relationships among the input fea-
tures. Recently, hardware improvements and very large
training datasets have allowed these deep learning techniques
to surpass other machine learning algorithms for many pro-
blems. In a famous and early example, scientists from Google
demonstrated that a neural network ‘discovered’ that cats,
faces and pedestrians were important components of online
videos [9] without being told to look for them. What if, more
rsif.royalsocietypublishing.org J. R. Soc. Interface 15: 20170387
2
on April 10, 2018http://rsif.royalsocietypublishing.org/Downloaded from
评论0