Computational Biology, Part 24 Clustering and Unmixing of Subcellular Patterns

Transcription

1 Computational Biology, Part 24 Clustering and Unmixing of Subcellular Patterns Robert F. Murphy Copyright 1996, 1999, All rights reserved.

2 Unsupervised Learning to Identify High-Resolution Protein Patterns

3 Location Proteomics Tag many proteins cdna tagging Put individual cdnas into GFP tagging vector (puts GFP coding at end) Transfect individual clones with each tagged cdna CD-tagging (developed by Jonathan Jarvik and Peter Berget): Infect population of cells with a retrovirus carrying DNA sequence that will tag in a random gene in each cell Isolate separate clones, each of which produces express one tagged protein Use RT-PCR to identify tagged gene in each clone Collect many live cell images for each clone using spinning disk confocal fluorescence microscopy or automated high-throughput microscopy

4 Images of CD-tagged 3T3 cells

5 Chen et al 2003; Chen and Murphy 2005 SLF features can be used to measure similarity of protein patterns This allows us for the first time to create a systematic, objective, framework for describing subcellular locations: a Subcellular Location Tree Start by grouping two proteins whose patterns are most similar, keep adding branches for less and less similar patterns

6 Protein name Human description From databases

7 Nucleolar Proteins

8 Punctate Nuclear Proteins

9 Predominantly Nuclear Proteins with Some Punctate Cytoplasmic Staining

10 Nuclear and Cytoplasmic Proteins with Some Punctate Staining

11 Uniform

12 Protein name Bottom: Visual Assignment to known locations Top: Automated Grouping and Assignment

13 Decomposing (unmixing) complex patterns

14 Decomposing mixture patterns Clustering or classifying whole cell patterns will consider each combination of two or more basic patterns as a unique new pattern Desirable to have a way to decompose mixtures instead One approach would be to assume that each basic pattern has a recognizable combination of different types of objects

15 Object type determination Rather than specifying object types, we can choose to learn them from the data Use subset of SLFs to describe objects Perform k-means clustering for k from 2 to 40 Evaluate goodness of clustering using Akaike Information Criterion Choose k that gives lowest AIC

16 Cluster Number Selection Akaike Information Criterion (AIC) = 2k 2ln(L) k=number of clusters L=likelihood of model given data 16

17 Example of Object Types Type A Type B Type C Type D 17

18 Unmixing: Learning strategy Once object types are known, each cell in the training (pure) set can be represented as a vector of the amount of fluorescence for each object type Learn probability model for these vectors for each class Mixed images can then be represented using mixture fractions times the probability distribution of objects for each class

19 Amt fluor. 0.2 Pure Lysosomal Pattern Object type Golgi class Lysosomal class Nuclear class Pure Golgi Pattern Amt fluor Object type Golgi class Lysosomal class Nuclear class 50% mix of each Amt fluor Object type All Golgi class Lysosomal class Nuclear class

20 Two-stage Strategy for unmixing unknown image Find objects in unknown (test) image, classify each object into one of the object types using learned object type classifier built with all objects from training images For each test image, make list of how often each object type is found Find the fractions of each class that give best match to this list

21 Test samples How do we test a subcellular pattern unmixing algorithm? Need images of known mixtures of pure patterns difficult to obtain naturally Created test set by mixing different proportions of two probes that localize to different cell parts (lysosomes and mitochondria)

22 Tao Peng, Ghislain Bonamy, Estelle Glory, Sumit Chanda, Dan Rines (Genome Research Institute of Novartis Foundation) Lysotracker

23 Mitotracker

24 Mixture of Lysotracker and Mitotracker

25 Pattern unmixing results 25