Data Scientist
Seattle,WA
thlindsay1 at gmail dot com
Machine Learning and Data Visualization
Excited by the opportunities and rapid pace of innovation in industry, I joined Insight Data Science as a fellow. Here I worked on a consulting project for Insight on a internal project where I developed a natural language search tool for their alumni slack network. This tool was built around a custom page-rank algorithm that I trained to identify question and answer sequences within the rather noisy conversations of the slack forums. You can read more about the approach I took here .
Along with my collaborators Floris Van Breugal and Peter Weir I worked on a project called figurefirst to enable agile construction of scientific figures. The main idea of figurefirst is to use the scalable vector graphics (SVG) language to specify a layout for a scientific figure. Figurefirst will parse this layout document and generate matplotlib plotting objects in python. This approach allows users to take advantage of SVG editors like inkscape to generate the layout and complex artwork, and then fuse these graphics with data-driven plotting code. Because the layout and plotting code can be developed independent of one another, it allows a much more iterative approach to figure creation where a scientist may start with a rough storyboard for a figure sequence using the layout, and then fill this out with data as analysis and experiments are completed.
Neuroscience Research
I completed my postdoctoral training in the Dickinson Lab at the California Institute of Technology. Broadly my neuroscience intrests focused on how a nervous system endows an animal with agency in the world around it. To generate behavior, an animal's nervous system must respond to incoming sensory information and respond with the appropriate motor acts. This is a challenging problem since it requires neural networks that can decode a constant stream of information arriving from an array of diverse sensors; transform this information into motor commands that will produce a desired response; and do this within the constraints placed by the physics of the body. Moreover, this operation must be updated in real time, since the movement itself will change what the animal senses. Finally, the operation should account for the internal desires and intention of the animal: is it searching for food? courting a mate? defending a resource? recovering from illness of injury? I view behavior as the result of a continuous loop in which reaction is fused with intention, transformed into action that updates sensation.
I studied the neural elements of this loop at several levels, in several different model systems. Following my undergraduate degree, I investigated how sensory neurons of mice responded to the injury of bone and pancreatic cancer. For my doctoral work, I studied how the simple neural network of the nematode C. elegans might process noxious stimuli while the worm searches for food.
For my postdoc work, looked at a small set of motor neurons that the frutfly uses to steer in flight. To crack this problem I developed a technique that allows us to optically record the activity of this motor network in intact flies as they fly in a virtual reality flight simulator. This research identifed a simple combinatorial strategy that the fly uses to adjust wing motion, indicating that the biomechanical properties of the wing joint are setup to give the nervous system a robust mechanical system to control. This has broader implications in the neural control of motion, as it provides an example of how muscle physiology and biomechanics may be adapted not simply to optimize power or force, but also the neural control of these outputs.
This shows the procedure I used to record synaptic currents from the locomotor command neurons of the nematode worm C. elegans. The video shows a recording from the AVA-neuron class which is involved in generating backward locomotion. The entire worm is about 1mm long, but the view in the video is focused on an area near the head where an important ganglion (collection of neurons) is located. In order to perform the procedure, the worms must first be fixed in place to a gelatinous film with some veterinary adhesive, and then immersed in a saline solution. Since these nematodes have a pressurized hydroskeleton we access the cells by making a very small puncture in the cuticle with a sharpened glass needle. If we're lucky, the internal pressure of the worm will push the neurons of interest out of the cuticle, exposing them to our recording electrode. We use transgenic worms that have been engineered to express a fluorescent protein in the target cells so we can visualize them under the microscope. To record the electrical potential inside the neuron we make use of the whole cell technique: we use a glass electrode, and form a tight seal between the neuron and the tip of the electrode. After forming this seal, a small bleb of the cell membrane appears inside the electrode. With a little suction we break this bleb and the saline inside the electrode becomes electrically contiguous with the voltage
inside the neuron
