Genetically identical cells and organisms raised in the same environment show considerable differences in phenotype. Single cell study of the operation of a signaling system in S. cerevisiae (eg Yu et al. 2008) enables quantification of sources of this variation (Colman-Lerner et al. 2005), and discovery of different means cells use to control signal to generate accurate quantitative responses given this variation (Brent 2009, Andrews et al. 2016). For example, one important source of phenotypic difference in individual cells and organisms is correlated variation (Colman-Lerner et al. 2005). Correlated variation in protein dosage is consequential: in S. cerevisiae and C. elegans, differences in global protein dosage operationally define different physiological states (Mendenhall et al. 2016) and cause differences in phenotypic expressivity and penetrance of mutant alleles. Deeper understanding of individual contributions to non-genetic variation (Pesce et al., 2018) will lead, among other outcomes, to better understanding of the effects of disease genes in the human population. In the shorter term, shallow understanding has enabled us to start building reduced-variation regulated controllers of gene expression to enable better experimentation in applied cancer research.
Our work draws on different kinds of knowledge, and we conduct it with the benefit of strong collaborations. Key collaborations today are with the Alejandro Colman-Lerner lab at the University of Buenos Aires, Fabian Rudolf, Joerg Stelling, and coworkers at ETHZ Basel, Laura Boucheron and coworkers at NMSU Las Cruces, Alexander Mendenhall and coworkers at UW, and emerging work with Bernease Herman, Valentina Staneva, and Ariel Rokem at the eScience institute at UW. Lab members are encouraged to participate in advisory and big picture government activities, and to think broadly about how ongoing increases in biological knowledge and capability (and their own research) might inflect the course of human affairs. To that end, we maintain an interaction with Alison Wylie (a feminist philosopher of science) and coworkers at UBC Vancouver.
22 March 2018
Pesce et al. Molecular and Systems Biology in press
In the yeast Pheromone Response System (PRS), mutants and other perturbations (for example Δbim1) that affect cytoplasmic microtubule + end function increase cell-cell variation in transmitted signal, and signaling is sometimes erratic.
In this work, we (Gustavo Pesce and many others) held stethoscope up to the outside of the machine when it was running and missing different bearings, and learned what we could from listening carefully to the noises following up carefully on the molecular cell biology cell biology.
Here, we learned a fair amount. Variability is caused by erratic signaling by a particular MAPK, yeast Fus3, when it is recruited to the membrane, at the signaling site.
A number of positive feedbacks that affect recruitment of proteins to the signaling site. Moreover, there are "cross stimulatory positive interactions" in which signaling site proteins stimulate recruitment of proteins to the site of polarized growth, and these proteins stimulate recruitment to the signaling site.
Given the above, our best guess is that small instabilities in signaling can become amplified into larger instabilities in recruitment of proteins to the signaling site, leading to larger instabilities in signaling.
We are also tempted to speculate that somehow competition for proteins in limited number causes the patch to reach a certain size... and then, something causes it to crash.
These malfunctions affect fate choices and cell polarity choices. Which would play out during embryonic development and in maintenance of the adult soma. How much of the burden of human disease might be due to small functional differences due to coding sequence polymorphisms in microtubule end function proteins? How much morbidity might be due to small quantitative functional differences in allelic forms present in the population, and could never be detected by blunt tools such as GWAS?
4 March 2018
First lab deep learning machine says Hello World
Last Wednesday BrentLab DL1 (Deep Learning 1; truename and ekename withheld to protect against remote enchantment) came alive. Built by Darren White downstairs from an old Dell Windows x86 box and motherboard, a cheap tower, and a 650W power supply from Amazon, and an Nvidia GTX 1080 Ti from Ebay in an auction at $240 over MSRP. In terms of FLOPS, BLDL1 is 24 X 10 exp 9 more powerful than the IBM 360/40 one lab member first worked with. BLDL1 plans to begin working hard helping pre-train a particular Convolutional Neural Network.