A Quick Guide for Developing Effective Bioinformatics Programming Skills

A Quick Guide to Organizing Computational Biology Projects

My Highlights from the meeting:

• Daphne Koller and Gal Chechik both spoke (Daphne as a keynote) on their recent work on identifying active motifs within networks (citation below).
• Uri Alon gave a beautiful keynote on how evolution for varying goals, when the goals share substructure, results in modularity.
• Aviv Bergman gave a nice keynote in a similar vein … on how evolution effects a network’s robustness, complexity, and the observed variance in the population.
• Moran Cabili gave a nice short talk on their recent publication on predicting tissue-specific metabolism (citation below).
• Amir Mitchell gave a nice talk on adaptive environmental conditioning.
• Todd Wasson gave a nice talk (and had a poster) on their model of competitive binding of DNA by nucleosomes and transcription factors. Tim Hughes, in the final keynote, presented a ton of data showing clear competition effects.
• There were four excellent talks on visualizing expression via imaging. Two were looking at the effects of enhacers (Eddy Rubin’s keynote and Jeffrey Chuang) and two focused on context-dependent expression in Drosophilia (Angela DePace and Erwin Frise).
• Robert Bradley had a very nice poster on his Fast Statistical Alignment (FSA) work. The ability to align multiple complete human genomes quickly and accurately.
• Michal Rabani had a poster on her recent work on identifying RNA motifs involved in post-transcriptional regulatory processes (citation below).

Gal Chechik, Eugene Oh, Oliver Rando, Jonathan Weissman, Aviv Regev, Daphne Koller (2008). Activity motifs reveal principles of timing in transcriptional control of the yeast metabolic network Nature Biotechnology DOI: 10.1038/nbt.1499

Tomer Shlomi, Moran N Cabili, Markus J Herrgård, Bernhard Ø Palsson, Eytan Ruppin (2008). Network-based prediction of human tissue-specific metabolism Nature Biotechnology, 26 (9), 1003-1010 DOI: 10.1038/nbt.1487

M. Rabani, M. Kertesz, E. Segal (2008). Computational prediction of RNA structural motifs involved in posttranscriptional regulatory processes Proceedings of the National Academy of Sciences, 105 (39), 14885-14890 DOI: 10.1073/pnas.0803169105

Discovering the underlying structure of a set of entities is a fundamental challenge for scientists and children alike. (Kemp PNAS)

The paper is focused on showing that discoveries about structural form can be understood computationally as probabilistic inferences about the organizing principles of a dataset.

Most structure-learning algorithms search for a structure of a single form that is assumed to be known in advance:

Clustering or competitive-learning algorithms search for a partition of the data into disjoint groups, algorithms for hierarchical clustering or phylogenetic reconstruction search for a tree structure, and algorithms for dimensionality reduction or multidimensional scaling search for a spatial representation of the data. (Kemp PNAS)

In the PNAS paper, they use graph grammars as a unifying language for expressing a wide range of structural forms.  This grammar provides a method for generating different kinds of structural forms (F).  Therefore, given a dataset (D) they can find the best form (F) and structure (S) of that form which best captures the relationships inherent between entities within the dataset P(S, F|D) ∝ P(D|S)P(S|F)P(F).   They use a uniform prior on P(F), a prior on P(S|F) which appropriately favors simpler forms, and constraints on P(D|S) which captures the intuitive expectation of data smoothness along the structure.    They then demonstrate that this model can accurately capture the intuitive structural form for a broad range of diverse datasets.

Holyoak, in a commentary on their PNAS paper, describe their work as “induction as model selection”.  It is therefore instructive to read Kemp and Tenenbaum’s Psychological Review paper (in press), where they discuss in some detail their perspective on inductive inference.  The Bayesian framework for probabilistic inference provides a general approach to understanding how problems of induction can, in principle, be solved.  Structured models are needed to capture the knowledge that drives induction.  Statistical inference is necessary to explain how this knowledge is acquired and used for inference in the presence of noise and uncertainty.  Bayesian methods therefore provide a principled framework for making “inherently under constrained inferences from impoverished data in an uncertain world.” (Griffiths, to appear)

In more detail, a learner must have both a recipe for specifying a prior and an engine for inductive inference.  The prior can be thought of simply as a priori knowledge about a problem.  Strong inductive bias is useful because it constrains the inferences a model is prepared to make.  (Constrains, however, only help if they guide in the right direction.)  Reasoning about a problem requires two kinds of knowledge.  First, one must have a structural representation which captures the kinds of relationships between entities, categories, or datum.   For example, a tree may be a good representation for evolution but a linear ordering may better describe body mass.  Note that enforcing a structure is equivalent to regularizing (or “cleaning up”) the data. Second, one must define a stochastic process to capture how the data depend on the structure.  Example processes include diffusion, drift, or causal transmission.  In other words, your process describes how you expect the data to change over the course of traversing your structure.

Kemp and Tenenbaum study high-level cognition, a field far from my own expertise.   While their efforts may not ground breaking to those in machine learning, their writing is remarkably readable, lucid, clear and thought provoking.

C. Kemp, J. B. Tenenbaum (2008). The discovery of structural form Proceedings of the National Academy of Sciences, 105 (31), 10687-10692 DOI: 10.1073/pnas.0802631105

K.J. Holyoak (2008). Induction as model selection. Proceedings of the National Academy of Sciences, 105 (31), 10687-10692 DOI: 10.1073/pnas.0805910105

C. Kemp, J. B. Tenenbaum (in press). Structured statistical models of inductive reasoning. Psychological Review. From Tenenbaum’s website.

T. L. Griffiths, C. Kemp, and J.B. Tenenbaum (to appear) Bayesian models of cognition. In Ron Sun (ed.), Cambridge Handbook of Computational Cognitive Modeling. Cambridge University Press. From Tenenbaum’s website.

Previous work from the O’Shea lab(Springer et. al. PLoS Biology 2003) on Pho4 indicated that the transcription factor, the major activator of the phosphate response pathway, has multiple phosphorylated states. In high phosphate conditions, Pho4 is fully phosphorylated, localized to the cytoplasm and therefore inactive. In phosphate starvation, Pho4 is completely unphosphorylated and fully active. The interesting case, however, is the intermediate state. When phosphate levels are intermediate, Pho4 is partially phosphorylated, accumulates in the nucleus and activates transcription of a subset of the phosphate-responsive genes. For example, Pho5 is a phosphate target which is activated only in the fully active (no phosphorylations) state whereas Pho84 is activated in both low and intermediate phosphorylation states. The differential affinity of Pho4 for some phosphate responsive genes was attributed to affinity differences.

A later paper by Buck and Lieb hinted at a possible mechanism behind the differential binding of Pho4. The Buck paper, studying an entirely different transcription factor (Rap1), identified differential binding in different glucose concentrations. Several nice experiments later and they propose that the mechanism of differential Rap1 binding is related to nucleosome positioning. In high glucose, nucleosomes are stabilized over potential Rap1 targets blocking them from the transcription factor whereas in low glucose the sites are accessible. Figure 6 from their Nature Genetics paper sums up the model:

Given that a great deal is known about nucleosome positioning effects at the Pho5 locus, a natural hypothesis is that chromatin may be an integral part of the affinity differences postulated by Springer et. al. This is precisely the hypothesis underlying the recent Lam et. al. paper (Nature 2008). To quote Lam, “We hypothesized that chromatin may influence gene expression by differentially regulating the accessibility of Pho4 sites int he Pho5 and Pho84 promoters.” They build a library of Pho5 promoter constructs, measure nucleosome positioning in these constructs and at other Pho genes, and demonstrate that, “the interplay of chromatin and binding-site affinity allows different promoters regulated by the same factor to interpret and respond to cellular signals uniquely.”

For may transcription factors, the motif which is recognized has been well characterized. These motifs, often represented as position-specific scoring matricies, show strong positive correlation with available protein-DNA data such as Chip-chip, Chip-seq, and classical promoter bashing. In other words, bound sites more often than not have the motif nearby. However, when viewed in the reverse direction — ie. scan a genome for a given motif, the data indicates that the transcription factors could potentially bind seemingly everywhere. Yet we know that binding is discriminant, specific, and yet not linearly related to site affinity. Including nucleosome positioning in models of transcriptional control may be a necessary component for understanding regulation.

Lam, F.H., Steger, D.J., O’Shea, E.K. (2008). Chromatin decouples promoter threshold from dynamic range. Nature, 453(7192), 246-250. DOI: 10.1038/nature06867

Buck, M.J., Lieb, J.D. (2006). A chromatin-mediated mechanism for specification of conditional transcription factor targets. Nature Genetics, 38(12), 1446-1451. DOI: 10.1038/ng1917

Springer, M., Wykoff, D.D., Miller, N., O’Shea, E.K. (2003). Partially Phosphorylated Pho4 Activates Transcription of a Subset of Phosphate-Responsive Genes. PLoS Biology, 1(2), e8. DOI: 10.1371/journal.pbio.0000028

I’ve been reading and thinking quite a bit lately about biological networks with respect to genetic interactions such as synthetic lethal (& synthetic sick), rescue, and dosage compensation. Oversimplifying, in all these cases the network is able to adjust to perturbation … sometimes in, as of yet, unpredictable fashion. The buzz words here are redundancy and robustness. A decent analogy already exist in this arena. Primarily associated with metabolic networks, the pipe analogy represents the pathways through the network as pipes which flow metabolites to reach some final destination. So a perturbation is effectively a disruption to a pipe. For robustness you depend on some other pipe in the system to pick up the load.

But lately I’ve been thinking of a different analogy. In this analogy the network is like a sports team. Pick your favorite sport as an example and follow along, but for now I’ll use (indoor) volleyball. Members of the team specialize in playing their particular position: a setter, a blocker, a libero (defense specialist), and hitters (attackers). You put six players on the court at any time, but have a bench full of alternative players. Traditionally your bench players are also somewhat specialized in their preferred roles but aren’t as good as your starters. Some of these alternate players are utilized for particular circumstances which arise frequently during the course of a game. Any and all players may be asked, in the course of a game, to perform duties outside their specialty, but rarely is a game decided by these plays.

So I hope you see the analogy. The players are components of the network and the function of the network is to win. Every component has a specialized function but is also likely capable of doing other functions (pleiotropy). Bench players are a form of redundancy but also specialization. You pick the components because they do particular jobs well and combinations of components because together they get the job done (winning). The ball, as it were, is now the information or stimulus which sets the whole gang into motion.

Let me stretch the analogy a bit further …. mutations (changes) come in two forms: players who improve (new workout or nutritional habits — rare) and those that get hurt (their game is “off”). But lets say, for the moment, that our star setter gets hurt so badly that she can’t even play (a deletion). Without the setter, the team can’t hold it together (she’s essential) and lose. Or, perhaps, the rest of the team may be able to compensate for her loss and still win (she was not essential) using different complements of players or slightly different strategies. These wins may be less convincing (reduced fitness), the team may present particular inadequacies which make it susceptible to specific attacks (environmental effects), or it may be very sensitive to the damage or loss of other players (synthetic lethals). Sometimes, rarely, the team discovers that it is actually better without the injured player (increased fitness).

Good sports teams are definitely organic in their abilities and play, but perhaps this analogy is too much of a stretch. Maybe it’s not that good of an analogy to begin with, and perhaps it isn’t “useful”. But it is at least timely, because of the current Olympics games.

Yu H. and Gerstein M. (2006). Genomic analysis of the hierarchical structure of regulatory networks. PNAS 103(40), 14724-14731. DOI: 10.1073/pnas.0508637103