Dynamic DNA nanotechnology from the test tube to the cell
We have developed systematic approaches for the design of molecular circuitry using DNA strand displacement (reviewed here). Starting in cell-free settings and with a limited but well understood toolbox of molecular interactions, we could rapidly scale up system complexity and quantitatively characterize reaction mechanisms to an extent that is not feasible for engineered gene circuits or other cell-based technologies. Using strand displacement as the basic engineering primitive we have created complex analog and digital circuits.. We're applying DNA strand displacement-based systems to develop molecular diagnostics such as sensors can be used to detect point mutations in single-stranded (here) and double-stranded (here) nucleic acids with a specificity comparable to the best enzyme-based methods. In recent work, we have also begun to show how DNA-based logic circuits can be adapted to the cellular environment (here). We are building on this initial success to create molecular control systems that can autonomously sense and respond to molecular cues in the cellular environment.
Programming gene expression
Controlling the timing and location of gene expression is desirable for a wide range of applications in biology and medicine. For example, an ideal cancer treatment combines strong therapeutic action in tumor cells with minimal side effects in surrounding healthy tissue. To achieve such specificity, we need to develop molecular control circuitry that can autonomously sense, analyze and modulate cell state. To this end, we are exploiting and rewiring existing biological pathways such as the mammalian microRNA pathway (here) or yeast signaling cascades (here).
Learning the rules of gene regulation from massively parallel reporter assays
A major hurdle for the de novo design of synthetic gene circuits is that we generally do not fully understand how the sequence of cis-regulatory elements --- promoters, enhancers, UTRs, splice regulatory regions or other --- relate to their function. In the absence of a well-characterized sequence-function relationship it is challenging to predictably modulate gene expression in synthetic constructs or to reliably predict how endogenous genes are expressed. In recent work, we showed that the grammar of biological cis-regulation can be learned from libraries for synthetic gene expression constructs with random sequence elements. For example, we have combined machine learning with RNA-seq of a splice reporter library with several million members to create a predictive model of alternative splicing (here). We then showed that this model could outperform existing models on the task of predicting the impact of human genomic variants on alternative splicing. More recently, we used a related approach to build a model that can predict how 5'UTR sequence determines proteins expression in yeast. We validated our model by engineering novel 5'UTR sequences with desired behaviors.
Technologies for single-cell sequencing
To forward engineer functional gene expression constructs that are targeted to certain cell types or states, it is first necessary to understand what defines such cell states. Single-cell RNA sequencing has emerged as maybe the most powerful tool for characterizing single cells at high throughput. However, existing scRNA-seq methods require complex fluidics and are costly which limits access to a relatively small number of labs. In recent work, we have developed a novel scRNA-seq technology, SPLiT-seq, a scRNA-seq method that labels the cellular origin of RNA through combinatorial indexing. SPLiT-seq is compatible with fixed cells, scales exponentially, uses only basic laboratory equipment, and costs one cent per cell.
We have developed systematic approaches for the design of molecular circuitry using DNA strand displacement (reviewed here). Starting in cell-free settings and with a limited but well understood toolbox of molecular interactions, we could rapidly scale up system complexity and quantitatively characterize reaction mechanisms to an extent that is not feasible for engineered gene circuits or other cell-based technologies. Using strand displacement as the basic engineering primitive we have created complex analog and digital circuits.. We're applying DNA strand displacement-based systems to develop molecular diagnostics such as sensors can be used to detect point mutations in single-stranded (here) and double-stranded (here) nucleic acids with a specificity comparable to the best enzyme-based methods. In recent work, we have also begun to show how DNA-based logic circuits can be adapted to the cellular environment (here). We are building on this initial success to create molecular control systems that can autonomously sense and respond to molecular cues in the cellular environment.
Programming gene expression
Controlling the timing and location of gene expression is desirable for a wide range of applications in biology and medicine. For example, an ideal cancer treatment combines strong therapeutic action in tumor cells with minimal side effects in surrounding healthy tissue. To achieve such specificity, we need to develop molecular control circuitry that can autonomously sense, analyze and modulate cell state. To this end, we are exploiting and rewiring existing biological pathways such as the mammalian microRNA pathway (here) or yeast signaling cascades (here).
Learning the rules of gene regulation from massively parallel reporter assays
A major hurdle for the de novo design of synthetic gene circuits is that we generally do not fully understand how the sequence of cis-regulatory elements --- promoters, enhancers, UTRs, splice regulatory regions or other --- relate to their function. In the absence of a well-characterized sequence-function relationship it is challenging to predictably modulate gene expression in synthetic constructs or to reliably predict how endogenous genes are expressed. In recent work, we showed that the grammar of biological cis-regulation can be learned from libraries for synthetic gene expression constructs with random sequence elements. For example, we have combined machine learning with RNA-seq of a splice reporter library with several million members to create a predictive model of alternative splicing (here). We then showed that this model could outperform existing models on the task of predicting the impact of human genomic variants on alternative splicing. More recently, we used a related approach to build a model that can predict how 5'UTR sequence determines proteins expression in yeast. We validated our model by engineering novel 5'UTR sequences with desired behaviors.
Technologies for single-cell sequencing
To forward engineer functional gene expression constructs that are targeted to certain cell types or states, it is first necessary to understand what defines such cell states. Single-cell RNA sequencing has emerged as maybe the most powerful tool for characterizing single cells at high throughput. However, existing scRNA-seq methods require complex fluidics and are costly which limits access to a relatively small number of labs. In recent work, we have developed a novel scRNA-seq technology, SPLiT-seq, a scRNA-seq method that labels the cellular origin of RNA through combinatorial indexing. SPLiT-seq is compatible with fixed cells, scales exponentially, uses only basic laboratory equipment, and costs one cent per cell.