Ãå±±ÂÖ¼é

On a circuit level, neurons code information based on their state and environmental context. Our lab goal is to develop an understanding of their governing principles, similar to the laws of other fields of science, such as physics and chemistry, in which distinct rules can be used to predict the outcome of a system. Knowledge of these principles will help to construct a basis for the development of biologically-informed treatment for addiction, PTSD, and a range of other psychiatric disorders. Also, this knowledge could inform artificial intelligence to develop natural intelligence based algorithms, which may result in new AI horizons.  For example, flies and mice can learn to navigate in a complex environment and make better and more efficient choices than the best algorithms running on powerful computers. Friedman lab will use its dual strength in computation and circuit physiology (Friedman et al., 2017; Friedman et al., 2015; Friedman et al., 2020; Friedman et al., 2016) to formulate these principles and generalize them to be useful by AI (Fig. 1). Even though these goals are ambitious, steps toward their accomplishment will be important for brain-physiology and AI communities. In the next paragraphs, we will explain our novel perspective to achieve this goal.

Also, there is a crucial need for biologically-informed, targeted treatments for a range of psychiatric disorders, including addiction and PTSD. Modern technology allows for simultaneous recording of hundreds of neurons, critical for identification of the mechanisms of behavior dysfunction. However, computational tools for analysis of simultaneously recorded big data sets have not yet emerged. Using our dual background in behavioral circuit physiology and computer science (Fig. 1), we will combine an experimental and computational approach to examine the role of the basal ganglia based circuit in addiction and PTSD and develop computational tools that elucidate abnormal circuit signal transduction.

In order to reach this ambitious goal, Friedman lab will have a fundamentally different strategy. We plan to use a dual approach to research that uses a computational perspective to guide experiments and experimentally obtained results to shape theories. Friedman lab work sits at the intersection between neuroscience and computation. Friedman lab major publications focus on brain circuit computation underlying dysfunction in addiction, depression, mood disorders and Huntington’s disease (Friedman et al., 2017; Friedman et al., 2015; Friedman et al., 2020; Friedman et al., 2016). To this end, Friedman lab passionate about studying brain disorders and diseases and developing experimental and computational approaches for evaluating large experimental datasets.

We developed computational approaches for analyses of my experimental work on dopamine dysfunction in depression and addiction. On the basis of this work, we went on to develop a patterned deep brain stimulation technique, which has been influential in the clinical application of DBS in addiction and depression (Friedman et al., 2009; Friedman et al., 2010). We studied cognitive and mood-related circuits of the brain’s cortico-striatal and dopamine systems. These two systems of the basal ganglia fundamentally subserve motor, mood and cognitive function and dysfunction. Recently, we publish 3 papers in Cell that describe an approach for studying cost-benefit decision-making to unpack the specialized microcircuitry between the prefrontal cortex and the striatum underlying a fundamental class of decisions (Friedman et al., 2017; Friedman et al., 2015; Friedman et al., 2020). For this, we have established new mathematical and computational approaches for clustering multidimensional data in a noisy environment (Friedman et al., 2015)  and for finding functional connectivity between processes with nonlinear interactions using multidimensional and dynamic hidden Markov models (Friedman et al., 2020; Friedman et al., 2016).

Our lab serves as a bridge to the experimental world of systems neuroscience that is experiencing a golden era of innovation and producing unprecedented levels of big data sets that require increasingly sophisticated computational approaches to analyze and evaluate. Our research program holds potential for improving understanding of brain computation, including in artificial intelligence, as well as improving the lives of those with brain disorders. We will perform experiments informed by computational modeling and further develop novel analytic techniques for interpreting complex data sets in new ways.

 In order to define brain principles and make them relevant to AI, and development biologically-informed and targeted treatments for a range of psychiatric disorders, Friedman lab will need to make major breakthroughs in 4 directions (Fig.1):

 First, we need to develop behavioral tasks that are relevant to decision-making and learning. These tasks need to be very simple and natural for the animal. However, regardless of simplicity, this task needs to involve neuronal circuit dynamics that occur during most learning or decision making processes. Behavior relies on multiple observed and hidden parameters that must be dissected in order to interpret measures of neuronal activity. Building on our current work, where we developed decision making task batteries (Friedman et al., 2015) and high-throughput apparatuses where dozens of rodents could learn tasks simultaneously (Friedman et al., 2020), we will develop behavioral task batteries that allow for precise interpretation of physiological recordings. This task also needs to be relevant to AI applications and relevant to study neurological and psychiatric disorders. So that information about each neuronal circuit’s dynamics at normal and disorder sates can be mimicked by algorithms.

Second, we need to record a ‘critical mass’ of neurons that is sufficient to decode neuronal activity. Recorded neurons need to be part of the same circuit which codes relevant information about the task. Therefore, to enable high throughput physiological recordings, we will develop methods to record connected neurons in multiple brain regions. Our research will build on our past work where we developed and integrated electrophysiology, imaging and perturbation methods that allow me to simultaneously record ~300-1000 neurons across multiple brain regions using electrophysiology or imaging. We will continue to refine this method to obtain higher density physiological recordings. Also, recordings need to be obtained in circuits that are relevant to learning and decision-making; importantly, circuits need to be well defined and simple enough that information obtained about circuit dynamics can be translated to biologically possible models and AI algorithms. Therefore, we will perform these recordings in cortico-striatal networks. Multiple cortical circuits code different informational aspects related to decision-making that are integrated by the striatum in a state-dependent manner. Striatal function is a cornerstone of motor, motivational and decision-making behavior; thus, the striatum is critically involved in learning, and its dysfunction is a common factor in a range of psychiatric and neurological disorders. Importantly, the striatum is also an attractive candidate for advanced computational work due to its well-defined anatomical organization and the complexity of intersecting cell type and compartment classifications. For example in our recent works (Friedman et al., 2017; Friedman et al., 2020), using Hodgkin Huxley and graph theory based models, we were able to recapitulate neuronal network dynamics in the striatum during decision-making and learning. These models can be relevant for development of AI algorithms.

Third, we will analyze the complex data that results from these high-density circuit recordings. There currently exist only a few bioinformatic methods suited for analyzing neuronal circuit dynamics when activity is simultaneously recorded across brain regions. Therefore, even with state-of-the-art analysis techniques, it is difficult to decode circuit activity. To fill this need, we will develop analytical computational tools capable of examining nonlinear connectivity of circuits, finding common patterns of circuit response, decoding circuit activity, and characterizing circuit dynamics. We will build off our previous work where we developed and applied neuroscience-specific computational techniques that allowed us to cluster signals from 10TB database of electrophysiological recordings (Friedman et al., 2015); find non-linear connectivity using multidimensional hidden Markov models (Friedman et al., 2016); find neuronal cells, spines and estimate a neuronal network from 30 TB of high resolution brain scans (Friedman et al., 2020); discover stages during cognitive discrimination learning and underlying it striatum dynamics  using a dynamic HMM algorithm (Friedman et al., 2020). We will capitalize on computational analysis tool development in order to analyze data obtained in my lab and provide a tool box for a community.

Fourth, we will develop modeling tools to extract, from experimental evidence across multiple sources, shared principles that may link circuit activity to specific behaviors. This modeling approach will be a stepping stone toward defining brain principles. To define these principles, Friedman lab will find and categorize similar network organizations, similar behavioral or internal states, and similar circuit dynamics patterns. We hypothesize that similarly organized networks during specific states will have the same activity dynamic patterns which govern information processing. We will build from the modeling approaches in our recent works (Friedman et al., 2017; Friedman et al., 2020). Using Hodgkin Huxley and graph theory models, we examined several network organizations that allowed us to predict the major effects of stress or Huntington’s disease on a cortico-striatal circuit and pinpoint what effects are critical for normal and aberrant learning and decision making. Our current models link multiple levels of information, including circuit anatomy, circuit activity and behavior (Friedman et al., 2020). In Friedman lab we will attempt to find models that can describe circuit activity across multiple behavioral tasks; ideally, these models will differentiate between circuit computational roles. These models will be critical for the development of AI algorithms, which will be inspired by the best striatal circuit models. We will use this modeling strategy to inform and guide subsequent experiments combining behavior and physiology. We will start my modeling development on striatal circuits, thereafter we will generalize them to other brain regions.

References:

Friedman, A., Frankel, M., Flaumenhaft, Y., Merenlender, A., Pinhasov, A., Feder, Y., Taler, M., Gil-Ad, I., Abeles, M., and Yadid, G. (2009). Programmed acute electrical stimulation of ventral tegmental area alleviates depressive-like behavior. Neuropsychopharmacology 34, 1057-1066.

Friedman, A., Homma, D., Bloem, B., Gibb, L.G., Amemori, K.I., Hu, D., Delcasso, S., Truong, T.F., Yang, J., Hood, A.S., et al. (2017). Chronic stress alters striosome-circuit dynamics, leading to aberrant decision-making. Cell 171, 1191-1205.

Friedman, A., Homma, D., Gibb, L.G., Amemori, K., Rubin, S.J., Hood, A.S., Riad, M.H., and Graybiel, A.M. (2015). A corticostriatal path targeting striosomes controls decision-making under conflict. Cell 161, 1320-1333.

Friedman, A., Hueske, E., Drammis, S.M., Toro Arana, S.E., Nelson, E.D., Carter, C.W., Delcasso, S., Rodriguez, R.X., Lutwak, H., DiMarco, K.S., et al. (2020). Striosomes Mediate Value-Based Learning Vulnerable in Age and Huntington’s Model. Cell In press.

Friedman, A., Lax, E., Dikshtein, Y., Abraham, L., Flaumenhaft, Y., Sudai, E., Ben-Tzion, M., Ami-Ad, L., Yaka, R., and Yadid, G. (2010). Electrical stimulation of the lateral habenula produces enduring inhibitory effect on cocaine seeking behavior. Neuropharmacology 59, 452-459.

Friedman, A., Slocum, J.F., Tyulmankov, D., Gibb, L.G., Altshuler, A., Ruangwises, S., Shi, Q., Toro Arana, S.E., Beck, D.W., Sholes, J.E., et al. (2016). Analysis of complex neural circuits with nonlinear multidimensional hidden state models. Proc Natl Acad Sci U S A 113, 6538-6543.