Convolutional LSTM Networks for Subcellular Localization of Proteins

1

Convolutional LSTM Networks for SubcellularLocalization of Proteins

Søren Kaae Sønderby, Casper Kaae Sønderby, Henrik Nielsen*, and Ole Winther

*Center for Biological Sequence AnalysisDepartment of Systems Biology

Technical University of Denmark

2

Protein sorting in eukaryotes

3

Feed-forward Neural Networks

Problems for sequence analysis:• No builtin concept of

sequence• No natural way of

handling sequences of varying length

• No mechanism for handling long range correlations (beyond input window size)

4

LSTM networks

xt: input at time tht-1: previous outputi : input gate, f : forget gate, o: output gate, g: input modulation gate, c: memory cell.

An LSTM (Long Short Term Memory) cell

The blue arrow head refers to ct−1.

LSTM networks• are easier to train than

other types of recurrent neural networks

• can process very long time lags of unknown size between important events

• are used in speech recognition, handwriting recognition, and machine translation

5

“Unrolled” LSTM network

Each square represents a layer of LSTM cells at a particular time (1, 2, ... t).

The target y is presented at the final timestep.

6

Regular LSTM networks

Bidirectional: one target per position

Double unidirectional: one target per sequence

7

Attention LSTM networks

Bidirectional, but with one target per sequence.

Align weights determine where in the sequence the network directs its attention.

8

Convolutional Neural Networks

A convolutional layer in a neural network consists of small neuron collections which look at small portions of the input image, called receptive fields.

Often used in image processing, where they can handle translation invariance.

First layer convolutional filters learned in an image processing network, note that many filters are edge detectors or color detectors

9

Our basic model

Conv.

LSTM

Conv.

LSTM

Conv.

LSTM

FFN

……t t+1 T

Target prediction at t=T

Softmax

xt xt+1 xT

Note that conv. weights are shared across sequence steps for the convolutional filters

1D convolution(variable width)

Y K P WAxtxt-1xt-2 xt+1 xt+2

Conv. weights

10

Conv.

LSTM

Conv.

LSTM

Conv.

LSTM……t T

xt xt+1 xT

Encoder

……ht ht+1 hT

Vectors containing the activations in each LSTM unit at each time step

Attention

Attention

Attention

Att. Weighting over sequence positions𝛼t 𝛼t+1 𝛼T

Decoder

t+1

Weighted hidden average

Softmax

Target prediction

FFN

Our model, with attention

11

Our model, specifications

– Input encoding: Sparse, BLOSUM80, HSDM and profile (R1×80)

– Conv. filter sizes: 1, 3, 5, 9, 15, 21 (10 of each)– LSTM layer: 1×200 units– Fully connected FFN layer: 1×200 units– Attention model: Wa (R200×400), va (R1×200)

12

MultiLoc architectureMultiLoc is an SVM-based based predictor using only sequence as input

13

MultiLoc2 architecture

MultiLoc2 corresponds to MultiLoc + PhyloLoc + GOLoc.

Thus, its input is not only sequence, but also metadata derived from homology searches.

14

SherLoc2 architecture

SherLoc2 corresponds to MultiLoc2 + EpiLoc

EpiLoc = a prediction system based on features derived from PubMed abstracts found through homology searches

15

Results: performance

16

Learned Convolutional Filters

17

Learned Attention Weights𝛼1 . . . . . . . . . 𝛼t . . . . . . . . . . 𝛼T

18

t-SNE plot of LSTM representation

19

Contributions1. We show that LSTM networks combined with convolutions

are efficient for predicting subcellular localization of proteins from sequence.

2. We show that convolutional filters can be used for amino acid sequence analysis and introduce a visualization technique.

3. We investigate an attention mechanism that lets us visualize where the LSTM network focuses.

4. We show that the LSTM network effectively extracts a fixed length representation of variable length proteins.

20

AcknowledgmentsThanks to:• Søren & Casper Kaae Sønderby,

for doing the actual implementation and training

• Ole Wintherfor supervising Søren & Casper

• Søren Brunakfor introducing me to the world of neural networks

• The organizersfor accepting our paper

• Youfor listening!