A lot of what I’ve worked on can be found by taking a look at my thesis -

Ennis M. (2023) Behavior quantification as the missing link between fields: Tools for digital psychiatry and their role in the future of neurobiology. PhD, Harvard University, Division of Medical Sciences. https://arxiv.org/abs/2305.15385

The various projects within that thesis are summarized below, along with a few details on my pre-thesis work and a preview of projects that are currently ongoing.

Summary

Links to a few other academic profiles:

Projects (and other experience)

RNNs of RNNs

Utilizing control theory tools, I worked on finding/proving novel stability theorems for multi-area recurrent neural networks (RNNs). My co-author and I then applied our theorems to improve RNN performance on sequential classification benchmarks. Stability is important for interpretability and safety of RNNs in sensitive use cases like medicine, and empirically it can allow vanilla RNNs to perform substantially better than typically possible. In my opinion, combining multi-area structural constraints with architecture constraints that facilitate stability (such as sparsity) has great potential for the revival of RNN relevance in machine learning at large.

You can find the code for our best performing (sequence processing) “RNN of RNNs” architecture here:
https://github.com/ennisthemennis/sparse-combo-net

Follow ups on this line of work remain ongoing, and I overview a number of such proposed future directions in my thesis. Our original publication can be found at (* = equal contribution):

L. Kozachkov* and M. Ennis*, and J.J. Slotine. RNNs of RNNs: Recursive Construction of Stable Assemblies of Recurrent Neural Networks. Advances in Neural Information Processing Systems, 2022. https://openreview.net/forum?id=2dgB38geVEU

Psychiatric patient daily audio diaries

A major argument of my thesis is that audio journals are uniquely well-suited to provide scientific opportunities for a wide range of psychiatry studies, yet at present they are largely ignored by both computational and traditional groups. Thus a key contribution of the thesis was convincing the speech sampling team of the NIMH’s AMPSCZ initiative to dedicate more resources towards the analysis of this datatype.

For the Baker Lab, I built software to synthesize data management, acoustic feature extraction, NLP, and visualization tools for the processing of daily patient audio diaries for research purposes. In my thesis I provided extensive background and justification for code design decisions, and characterized pipeline outputs on a large lab dataset collected from individuals with Bipolar disorder.

You can find my pipeline here:
https://github.com/dptools/process_audio_diary

Notably, pilot scientific results from my work supported many of the theoretical arguments made for the utility of audio journals. They are highly versatile in their abilities, capable of simultaneously bringing data science tractability to more traditional psychiatric speech sample studies and bringing qualitative clinically interesting insights to studies centered on processing dense data streams. A speech sampling study could feasibly perform an exploratory analysis with actual statistical power if audio journals were used, while a passive sensing digital psychiatry study would be much more equipped to generate interesting follow-up hypotheses with access to journals. Quantitative and qualitative work need not be mutually exclusive, nor do objective and subjective metrics. The daily audio diary format can be well utilized in all these ways, and it can be well utilized in conjunction with many different datatypes or even entirely on its own.

Software infrastructure for the NIH’s AMPSCZ project
(clinical interview recordings)

I wrote code for data organization and quality control of psychiatric research interview audio and video recordings for AMPSCZ, a $100M+ NIMH initiative to identify predictors for future Schizophrenia diagnosis in youth at clinical high risk for psychosis. A wide variety of multimodal datatypes are being collected longitudinally from patients and controls across nearly 40 sites globally. To facilitate quality data collection in such a large and diverse collaborative project requires more thoughtful software infrastructure than a typical academic project. Handling audio/video recordings of patients brings additional privacy concerns as well.

You can find my pipeline here: https://github.com/dptools/process_offsite_audio

Early results from AMPSCZ data collection are described in my thesis, along with background on interview recording datatypes.

The thesis also includes scientific results from a Bipolar disorder clinical interview dataset collected by my lab, adapted from:

E. Liebenthal, M. Ennis, H. Rahimi-Eichi, E. Lin, Y. Chung, and J.T. Baker. Linguistic and non-linguistic markers of disorganization in psychotic illness. Schizophrenia Research, 2022. https://doi.org/10.1016/j.schres.2022.12.003

Passive sensing technology for measuring OCD

In chapter 3 of my thesis, I presented results from a multimodal digital psychiatry dataset collected during a deep brain stimulation (DBS) trial, showing detection of salient behaviors that would have otherwise gone unnoticed – a first of its kind blueprint for future research to connect computational psychiatry and neurobiology. In 2020, I received a Society of Biological Psychiatry Travel Award to present this work at SOBP’s annual meeting.

Though the meeting ended up being canceled because of COVID, I finished my analyses on the dataset for my thesis, and a corresponding manuscript is now in preparation. Additionally, the DBS trial that we collected data from was itself a novel potential treatment paradigm; the clinical results from this pilot “n of 1” trial are detailed in:

S.T. Olsen, I. Basu, M.T. Bilge, A. Kanabar, M.J. Boggess, A.P. Rockhill, A.K. Gosai, E. Hahn, N. Peled, M. Ennis, I. Shiff, K. Fairbank-Haynes, J.D. Salvi, C. Cusin, T. Deckersbach, Z. Williams, J.T. Baker, D.D. Dougherty, and A.S. Widge. Case report of dual-site neurostimulation and chronic recording of cortico-striatal circuitry in a patient with treatment refractory obsessive compulsive disorder. Frontiers in Human Neuroscience, 2020. https://doi.org/10.3389/fnhum.2020.569973

School work and personal projects

Previous internship experience includes quant finance, sports analytics, computational biology, software engineering, and biology bench work. Some specific projects are:

  • With the Jacksonville Jaguars (summer 2017), I worked with different player evaluation tools to identify strengths and limitations of existing scores used by the front office.

  • At Novartis Institutes for BioMedical Research (summer 2016), I developed a novel technique to link genome wide association study results and epigenetic data, investigating potential functional roles of genetic risk factors for Schizophrenia.

  • At Google (summer 2015), I implemented a distributed stats computation tool to evaluate product recommendations for dynamic display ads, including integrating my code into production systems. I was also an intern there in summer 2014, when I did front end software engineering work on the Google Drive mobile website.

I also had a ton of fun at Jane Street in summer 2022, and will be returning there for trading full time in 2024.

Internships

I love sharing things that I’m excited about, so I’ve made a point to spend some time teaching and mentoring even though my PhD program did not require it. I especially enjoy helping out in an office hours style setting, which was evident in my course evaluations too.

During my M.Eng year at MIT, I spent significant time on my role as a teaching assistant, which also provided full tuition coverage and stipend for my program. In fall 2016 I TAed 6.034 (Intro to Artificial Intelligence) and in spring 2017 I TAed 6.004 (Computation Structures). I had previously done volunteer teaching in undergrad, such as co-leading a January term primer on algorithms and designing/running a summer seminar for high school students through MIT’s HSSP, but this was my first in-depth experience with managing a classroom. In fall 2019 I was fortunate to have the opportunity to further improve my teaching skills, when I assisted with organizing/developing a new MIT course “Technologies for Mental Health and Wellness”.

In my first year at Harvard I participated in the medical school’s HPREP program, where I mentored a student both on learning biomedical concepts and writing applications for summer internships and eventually college. Once established in my lab, I supervised extended research projects for two different high school students - one who worked on developing algorithms for analysis of the lab’s wrist accelerometry data, and another who I guided in designing their own data science project in a subject of their choosing. I greatly enjoyed the mentoring process, and the students that I worked with went on to attend Duke, Princeton, and Columbia.

Teaching and mentoring

In the first year of graduate school, I rotated in a few different labs and gained some insight into a variety of research topics:

  • In the Ginty Lab I studied touch sensory processing in mouse models of Autism.

  • In the Carlezon Lab I studied ultrasonic vocalizations made by mice when under distress.

  • In the Macosko Lab I studied single cell RNA sequencing results from postmortem human brain samples.

Prior to beginning my PhD, I helped out with a number of research projects on a part time basis. For my M.Eng thesis, I worked in the Fee lab on unsupervised learning techniques to better understand features of Zebra finch song across development.

Ennis M. (2017) Unsupervised learning to quantify differences in song learning of experimental zebra finch populations. M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science. https://dspace.mit.edu/handle/1721.1/119521

My primary undergraduate research work was with the Feng Lab, working on characterizing a novel Williams syndrome mouse model.

I frequently helped around the lab with other projects as well, including assisting with an optogenetics setup for an Autism social motivation experiment, and scoring mouse behavior for a novel Schizophrenia model. I was acknowledged in “Direct modulation of GFAP-expressing glia in the arcuate nucleus bi-directionally regulates feeding” (eLife 10/2016), and my main project with the lab was eventually published as:

Barak B, Zhang Z, Liu Y, Nir A, Trangle SS, Ennis M, Levandowski KM, Wang D, Quast K, Boulting GL, Li Y, Bayarsaihan D, He Z, Feng G. (2019) Neuronal deletion of Gtf2i, associated with Williams syndrome, causes behavioral and myelin alterations rescuable by a remyelinating drug. Nature Neurosci. 22(5):700-708. PMID: 31011227.

During undergrad I also did a number of extensive class projects, one of which (on computational models of Schizophrenia) resulted in a new project in Patrick Winston’s Genesis AI Group. By discussing ideas with various PIs during PhD interviews, I made a variety of connections, and ended up receiving a NeurIPS 2019 travel award despite not having a paper at the conference.

Acknowledgements I received from this time include:

  • Inducing Schizophrenia in an Artificially Intelligent Story-Understanding System (ACS/Cognitive Systems Foundation, 8/2018)

  • Is Smaller Better? A Proposal to Use Bacteria For Neuroscientific Modeling (Frontiers in Computational Neuroscience, 2/2018)

  • The Deep Learning Revolution (MIT Press, 10/2018)

In high school, I was fortunate to have the opportunity to spend my summers at The Rockefeller University. I learned patch clamp techniques in the Pfaff Lab (2012) and examined the effects of obesity on immune cell composition in the human colon with the Breslow Lab (2010/2011).

With the Breslow Lab, I was also able to attend Digestive Disease Week in 2011 to help present some of my work, published as the following abstracts:

Pendyala S, Ennis MM, Fuentes-Duculan J, Holt PR. (2011) Weight Loss in Obesity Reduces Colorectal Inflammation With Upregulation of FOXP3 Regulatory T Cells. Digestive Disease Week.

Pendyala S, Ennis MM, Fuentes-Duculan J, Holt PR. (2011) Obese Compared to Normal Weight Subjects Show More Inflammatory Markers but Reduced FOXP3 Regulatory T Cells in the Colorectal Mucosa. Digestive Disease Week.

Research collaborations and volunteering

As a double major in Electrical Engineering and Computer Science (EECS) and Brain and Cognitive Sciences (BCS) with a minor in Mathematics at MIT, I took a wide variety of courses — and I continued that tradition throughout my graduate studies.

Relevant advanced coursework is as follows; for a full account, see my MIT transcript and Harvard transcript.

  • Computer Science and Engineering - Computation Structures, Design and Analysis of Algorithms, Software Construction, Signals and Systems, Artificial Intelligence, Theory of Computation, Statistical Learning Theory, Nonlinear Control

  • Mathematics - Multivariable Calculus, Differential Equations, Linear Algebra, Discrete Math, Probability and Random Variables, Real Analysis, Topology, Numerical Methods

  • Psychology and Neurobiology - Cognitive Processes, Animal Behavior, Infant and Childhood Cognition, Systems Neuroscience, Neurobiology of Psychiatric Disease, Comparative Neuroanatomy, Neuroimmunology

  • Computational Neuroscience - Neural Computation, Computational Cognitive Science, Principles of Neuroengineering, Cellular Biophysics, Quantitative Systems Physiology, Deep Learning in the Life Sciences, Technologies for Mental Health and Wellness

Additionally, I received the following academic awards at MIT:

  • 2015 Walle J. H. Nauta Award for Outstanding Research in BCS

  • 2016 Hans Lukas Teuber Award for Outstanding Academics in BCS

  • 2016 Eta Kappa Nu EECS Honor Society (top third of EECS majors at MIT)

  • 2016 Phi Beta Kappa Academic Honor Society (~75 seniors chosen each year at MIT for excellence in liberal arts)

  • 2016 Tau Beta Pi Engineering Honor Society (top 20% of engineering majors at MIT)

In high school, I got a shockingly comprehensive education on doing independent research: from designing biology lab projects to ordering reagents to carrying out experiments to managing younger students to even running a journal club.

Typically one forgets about high school entirely on a portfolio site like this, but I think it is a credit to the quality of education I was lucky to receive, so here’s a bit more on what I did at Pingry -

  • 2009 Waksman Student Scholar, which involved both an immersive summer program and bringing the project back to my school during my sophomore academic year. We posted many gene sequences from Duckweed to NCBI’s database.

  • 2010 Bio Olympiad semifinalist.

  • 2011 Siemens Competition semifinalist.

  • 2012 winner of the Dana Foundation’s “Design a Brain Experiment” contest.

  • Student lead of the independent research team (iRT) and associated journal club throughout my time at Pingry, which was deeply educational despite the obvious futility of my attempts to study Huntington’s Disease in zebrafish in the high school’s lab.

During my free time, I’m still making an effort to go through new coursework for fun, from an abstract algebra text to Duolingo’s Hungarian course. I also like to work on mini home improvement projects and painting 3D printed art.

In development

  • Adapting the concluding chapter of my thesis into a perspective piece to submit for publication, in addition to wrapping up the publication of results of the DBS case report. I will also be writing some blog posts for this website that summarize other snippets of my thesis in an approachable way.

    Academic Writing

  • Building software for processing of the audio journals being collected by AMPSCZ, and preparing official publication(s) related to my work on this datatype for Baker Lab. I will remain a consultant for the group through fall 2023 to ensure these projects are wrapped up appropriately. The code in progress for AMPSCZ diaries, nearing completion and currently running for a handful of test sites, is linked.

    Digital Psychiatry Research (https://github.com/dptools/daily_journal_dataflow_qc)

  • More deeply characterizing our architecture from the "RNNs of RNNs" paper in an empirical context, and working on various ideas for further performance improvements as well as better scalability and tools for interpretability. I plan to remain affiliated with the nonlinear systems lab at MIT through 2023. An initial preprint from this follow-up work is linked.

    Recurrent Neural Networks Research (https://arxiv.org/abs/2310.01571)