Harish Chandran
At Alphabet, I currently lead Behavior Models group within Waymo which is responsible to improve the quality of predictions about the behavior of agents the Waymo driven encounters in the real world. Previously, I worked on bring DeepMind technologies to Google products and making ranking and recommendations of posts on Google+ better. In graduate school, I worked on DNA self-assembly, nanoscience and theoretical self-assembly.
Research Areas
Authored Publications
Sort By
Balance Regularized Neural Network Models for Causal Effect Estimation
Mehrdad Farajtabar
Andrew Lee
Yuanjian Feng
Vishal Gupta
Peter Dolan
Martin Szummer
Causal Discovery & Causality-Inspired Machine Learning Workshop, NeurIPS 2020 (2020)
Preview abstract
Estimating individual and average treatment effects from observational data is an important problem in many domains such as healthcare and e-commerce. In this paper, we advocate balance regularization of multi-head neural network architectures. Our work is motivated by representation learning techniques to reduce differences between treated and untreated distributions that potentially arise due to confounding factors. We further regularize the model by encouraging it to predict control outcomes for individuals in the treatment group that are similar to control outcomes in the control group. We empirically study the bias-variance trade-off between different weightings of the regularizers, as well as between inductive and transductive inference.
View details
Improving the Performance of DNA Strand Displacement Circuits by Shadow Cancellation
Tianqi Song
Nikhil Gopalkrishnan
Abeer Eshra
Sudhanshu Garg
Reem Mokhtar
Hieu Bui
John Reif
ACS Nano (2018)
Preview abstract
DNA strand displacement circuits are powerful tools that can be rationally engineered to implement molecular computing tasks because they are programmable, cheap, robust and predictable. A key feature of these circuits is the use of catalytic gates to amplify signal. Catalytic gates tend to leak, that is, they generate output signal even in the absence of intended input. Leaks are harmful to the performance and correct operation of DNA strand displacement circuits. Here, we present "shadow cancellation", a general-purpose technique to mitigate leak in catalytic DNA strand displacement circuits. Shadow cancellation involves constructing a parallel shadow circuit that mimics the primary circuit and has the same leak characteristics. It is situated in the same test tube as the primary circuit and produces "anti-background" DNA strands that cancel "background" DNA strands produced by leak. We demonstrate the feasibility and strength of the shadow leak cancellation approach through a challenging test case, a cross-catalytic feedback DNA amplifier circuit that leaks prodigiously. Shadow cancellation dramatically reduced the leak of this circuit and improved the signal-to-background difference by several folds. Unlike existing techniques, it makes no modifications to the underlying amplifier circuit and is agnostic to its leak mechanism. Shadow cancellation also showed good robustness to concentration errors in multiple scenarios. This work introduces a direction in leak reduction techniques for DNA strand displacement amplifier circuits, and can potentially be extended to other molecular amplifiers.
View details
Modeling DNA Nanodevices Using Graph Rewrite Systems
Reem Mokhtar
Sudhanshu Garg
Hieu Bui
Tianqi Song
John Reif
Springer International Publishing (2017), pp. 347-395
Preview abstract
DNA based nanostructures and devices are becoming ubiquitous in nanotechnology with rapid advancements in theory and experiments in DNA self-assembly which have led to a myriad of DNA nanodevices. However, the modeling methods used by researchers in the field for design and analysis of DNA nanostructures and nanodevices have not progressed at the same rate. Specifically, there does not exist a formal system that can capture the spectrum of the most frequently intended chemical reactions on DNA nanostructures and nanodevices which have branched and pseudo-knotted structures. In this paper we introduce a graph rewriting system for modeling DNA nanodevices. We define pseudo-DNA nanostructures (PDNs), which describe the sequence information and secondary structure of DNA nanostructures, but exclude modeling of tertiary structures. We define a class of labeled graphs called DNA graphs, that provide a graph theoretic representation of PDNs. We introduce a set of graph rewrite rules that operate on DNA graphs. Our DNA graphs and graph rewrite rules provide a powerful and expressive way to model DNA nanostructures and their reactions. These rewrite rules model most conventional reactions on DNA nanostructures, which include hybridization, dehybridization, base-stacking, and a large family of enzymatic reactions. A subset of these rewrite rules would likely be used for a basic graph rewrite system modeling most DNA devices, which use just DNA hybridization reactions, whereas other of our rewrite rules could be incorporated as needed for DNA devices for example enzymic reactions. To ensure consistency of our systems, we define a subset of DNA graphs which we call well-formed DNA graphs, whose strands have consistent 5' to 3' polarity. We show that if we start with an input set of well-formed DNA graphs, our rewrite rules produce only well-formed DNA graphs.
View details
Probabilistic Analysis of Localized DNA Hybridization Circuits
Neil Dalchau
Nikhil Gopalkrishnan
Andrew Phillips
John Reif
ACS Synthetic Biology (2015)
Preview abstract
Molecular devices made of nucleic acids can perform complex information processing tasks at the nanoscale, with potential applications in biofabrication and smart therapeutics. However, limitations in the speed and scalability of such devices in a well-mixed setting can significantly affect their performance. In this paper, we propose designs for localized circuits involving DNA molecules that are arranged on addressable substrates and interact via hybridization reactions. We propose designs for localized elementary logic circuits, which we compose to produce more complex devices, including a circuit for computing the square root of a four bit number. We develop an efficient method for probabilistic model-checking of localized circuits, which we implement within the Visual DSD design tool. We use this method to prove the correctness of our circuits with respect to their functional specifications, and to analyze their performance over a broad range of local rate parameters. Specifically, we analyze the extent to which our localized designs can overcome the limitations of well-mixed circuits, with respect to speed and scalability. To provide an estimate of local rate parameters, we propose a biophysical model of localized hybridization. Finally, we use our analysis to identify constraints in the rate parameters that enable localized circuits to retain their advantages in the presence of unintended interferences between strands.
View details
Preview abstract
The tile assembly model is a Turing universal model of self-assembly where a set of square shaped tiles with programmable sticky sides undergo coordinated self-assembly to form arbitrary shapes, thereby computing arbitrary functions. Activatable tiles are a theoretical extension to the Tile assembly model that enhances its robustness by protecting the sticky sides of tiles until a tile is partially incorporated into a growing assembly. In this article, we experimentally demonstrate a simplified version of the Activatable tile assembly model. In particular, we demonstrate the simultaneous assembly of protected DNA tiles where a set of inert tiles are activated via a DNA polymerase to undergo linear assembly. We then demonstrate stepwise activated assembly where a set of inert tiles are activated sequentially one after another as a result of attachment to a growing 1-D assembly. We hope that these results will pave the way for more sophisticated demonstrations of activated assemblies.
View details
DNA Computing
Preview
Hieu Bui
Sudhanshu Garg
Nikhil Gopalkrishnan
Reem Mokhtar
John H. Reif
Tianqi Song
Computing Handbook (2014)
Meta-DNA: A DNA-Based Approach to Synthetic Biology
Nikhil Gopalkrishnan
Bernard Yurke
John H. Reif
Systems and Synthetic Biology: A Systematic Approach (2014)
Preview abstract
The goal of synthetic biology is to design and assemble synthetic systems that mimic biological systems. One of the most fundamental challenges in synthetic biology is to synthesize artificial biochemical systems, which we will call meta-biochemical systems, that provide the same functionality as biological nucleic acids-enzyme systems, but that use a very limited number of types of molecules. The motivation for developing such synthetic biology systems is to enable a better understanding of the basic processes of natural biology, and also to enable re-engineering and programmability of synthetic versions of biological systems. One of the key aspects of modern nucleic acid biochemistry is its extensive use of protein enzymes that were originally evolved in cells to manipulate nucleic acids, and then later adapted by man for laboratory use. This practice provided powerful tools for manipulating nucleic acids, but also limited the extent of the programmability of the available chemistry for manipulating nucleic acids, since it is very difficult to predictively modify the behavior of protein enzymes. Meta-biochemical systems offer the possible advantage of being far easier to re-engineer and program for desired functionality. The approach taken here is to develop a biochemical system which we call meta-DNA (abbreviated as mDNA),Meta-DNA (mDNA) based entirely on strands of DNA as the only component molecules. Our work leverages prior work on the development of self-assembled DNA nanostructures. Each base of a mDNA Meta-DNA (mDNA)is a DNA nanostructure. Our mDNA bases are paired similar to DNA bases, but have a much larger alphabet of bases, thereby providing increased power of base addressability. Our mDNA bases can be assembled to form flexible linear assemblies (single stranded mDNA) analogous to single stranded DNA, and can be hybridized to form stiff helical structures (duplex mDNA) analogous to double Double strand meta-DNA (dsmDNA) stranded DNA, and also can be denatured back to single stranded mDNA. Our work also leverages the abstract activatable tile model developed by Majumder et al. and prior work on the development of enzyme-free isothermal protocols based on DNA hybridization and sophisticated strand displacement hybridization reactions. We describe various isothermal hybridization reactions that manipulate our mDNA in powerful ways analogous to DNA–DNA reactions and the action of various enzymes on DNA. These operations on mDNA include (i) hybridization of single strand mDNA (ssmDNA)Single strand meta-DNA (ssmDNA) into a double strand mDNA (dsmDNA)Double strand meta-DNA (dsmDNA) and heat denaturation of a dsmDNA Double strand meta-DNA (dsmDNA)into its component ssmDNA Single strand meta-DNA (ssmDNA)(analogous to DNA hybridization and denaturation), (ii) strand displacement of one ssmDNA Single strand meta-DNA (ssmDNA)by another (similar to strand displacement in DNA), (iii) restriction cuts on the backbones of ssmDNA Single strand meta-DNA (ssmDNA)and dsmDNA Double strand meta-DNA (dsmDNA)(similar to the action of restriction enzymes on DNA), (iv) polymerization chain reactions that extend ssmDNA Single strand meta-DNA (ssmDNA)on a template to form a complete dsmDNA Double strand meta-DNA (dsmDNA)(similar to the action of polymerase enzyme on DNA), (v) isothermal denaturation of a dsmDNA Double strand meta-DNA (dsmDNA)into its component ssmDNA Single strand meta-DNA (ssmDNA)(like the activity of helicase enzyme on DNA) and (vi) an isothermal replicator reaction which exponentially amplifies ssmDNA Single strand meta-DNA (ssmDNA)strands (similar to the isothermal PCR reaction). We provide a formal model to describe the required properties and operations of our mDNA, and show that our proposed DNA nanostructures and hybridization reactions provide these properties and functionality.
View details
Preview abstract
This chapter overviews the current state of the emerging discipline of DNA nanorobotics that make use of synthetic DNA to self-assemble operational molecular-scale devices. Recently there have been a series of quite astonishing experimental results—which have taken the technology from a state of intriguing possibilities into demonstrated capabilities of quickly increasing scale and complexity. We first state the challenges in molecular robotics and discuss why DNA as a nanoconstruction material is ideally suited to overcome these. We then review the design and demonstration of a wide range of molecular-scale devices; from DNA nanomachines that change conformation in response to their environment to DNA walkers that can be programmed to walk along predefined paths on nanostructures while carrying cargo or performing computations, to tweezers that can repeatedly switch states. We conclude by listing major challenges in the field along with some possible future directions.
View details