Mila-IBM Research Project Proposal Draft

--- tags: past --- # Mila-IBM Research Project Proposal Draft - Highlight the important components so that I could then write it - Detail of the research challenge and desired impact - Literature section (maybe needed) **Motivation:** AI is being useful in prediction/classfication in biology, supervised learning approaches can discriminate states but have no concept of interaction. We are interested in helping experimentation and discovery and answer questions, such as "how can we try" to revert diseases, or mantain a system healthy. A (un)supervised learning model can passively classify the types of antibodies, or the various cell types, this passivity is a core assumption. In Optimal Control and Reinforcement Learning decision making is the first class citizen and prediction a tool to take informed decisions. (some literature about control and biological data: Mahdi Imani at al. Optimal Control of Gene Regulatory Networks with Unknown Cost Function) **Starting point:** The starting point of our work is based on [SERGIO](https://www.sciencedirect.com/science/article/pii/S2405471220302878), which is state-of-art for simulating transcriptomics data. In computational biology, one of the biggest problem is the amount of the data one has access to. This is also one crucial reason why there has not been much research about such application. We overcome this by making use of SERGIO to have unlimited amount of in silico transcriptomics data and known dynamics to optimize a control model. **The biological settings:** The focus will be considering transcritomics public datasets, such as https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE147844, which contains cell trajectories from healthy cells to unhealthy over a period of 2 weeks. And the problem is making controlling a cell fate, from unhealthy to healthy. This data comes from mice. We can also consider the c.elegans transcriptomics as the literature on c.elegans is more extensive than the mouse and the biological system is several order of magnitude simpler than the mouse. We want to show that we can intervene in a data driven way to control cells at the molecular level. We model the system with stochastic differential equations(SDEs) well understood in the transcriptomics field, and then we look at how well our optimal control algorithms can guide the system to healthy states. This will allow us to improve the biological models, leveraging the flexibility of deep learning models augmenting the current SDEs by modeling the known unknowns with residual models https://arxiv.org/abs/2003.10775. **The goal:** We are interestd in deriving insights from control methods in the bioloigcal settings and in answering questions such as can we keep a system healthy or revert it back to when it was healthy?. We know the community is focused on understanding the data itself, but using control models it will help us understands unknown factors, such as, which are the main dyanimics in the cell for a given state. For instance knowing that are some curcial genes or markers that are always "controlled" in order to have the desired state is very insightful. This has a lot of potential for the future of molecular biologist and medicine. **Literature**: - [SERGIO: A Single-Cell Expression Simulator Guided by Gene Regulatory Networks (2020)](https://www.sciencedirect.com/science/article/pii/S2405471220302878) - [Generalizing RNA velocity to transient cell states through dynamical modeling, Bergen et al.,2020, Nature Biotechnology](https://www.nature.com/articles/s41587-020-0591-3) - [SCENIC: single-cell regulatory network inference and clustering](https://www.nature.com/articles/nmeth.4463) - [Simulating multiple faceted variability in single cell RNA sequencing](https://www.nature.com/articles/s41467-019-10500-w.pdf) - [GeneNetWeaver: In silico benchmark generation and performance profiling of network inference methods](https://www.researchgate.net/publication/51242143_GeneNetWeaver_In_silico_benchmark_generation_and_performance_profiling_of_network_inference_methods) - [Understanding how T helper cells learn to coordinate effective immune responses through the lens of reinforcement learning](https://arxiv.org/pdf/1904.05581.pdf) - [Is T Cell Negative Selection a Learning Algorithm?](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7140671/) - [DeepCpG: accurate prediction of single-cell DNA methylation states using deep learning (EBI 2017)](https://genomebiology.biomedcentral.com/articles/10.1186/s13059-017-1189-z) - [Deep reinforcement learning for de novo drug design](https://advances.sciencemag.org/content/4/7/eaap7885?intcmp=trendmd-adv) - [Optimality principles in sensorimotor control](https://www.nature.com/articles/nn1309) - [Splatter: simulation of single-cell RNA sequencing data](https://genomebiology.biomedcentral.com/articles/10.1186/s13059-017-1305-0) - [Xiuwei et al 2019 - Simulating multiple faceted variability in single cell RNA sequencing](https://www.nature.com/articles/s41467-019-10500-w) - [Reinforcement Learning and Episodic Memory in Humans and Animals: An Integrative Framework](https://www.annualreviews.org/doi/abs/10.1146/annurev-psych-122414-033625) - [Multiscale relevance and informative encoding in neuronal spike trains](https://link.springer.com/article/10.1007/s10827-020-00740-x) - [Statistical criticality arises in most informative representations](https://iopscience.iop.org/article/10.1088/1742-5468/ab16c8) - [Epigenetic regulation of brain region-specific microglia clearance activity](https://www.nature.com/articles/s41593-018-0192-3) - [Epigenetics Control Microglia Plasticity - review of the one above](https://www.frontiersin.org/articles/10.3389/fncel.2018.00243/full) ### Literature on optimal control (related to bio) - [Optimal Control of Gene Regulatory Networks with Unknown Cost Function](https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8431514) - [Control of Gene Regulatory Networks With Noisy Measurements and Uncertain Inputs](https://ieeexplore.ieee.org/abstract/document/8017565?casa_token=bu2XloeveT0AAAAA:3WIhdmp8OKrr8B3QyH92sxIjF7pZGiwJaXe8DJvotCsFmghRPtodoxJj5Ta81kcFbJkYUb_8-aj1) - [Control of Gene Regulatory Networks Using Bayesian Inverse Reinforcement Learning](https://ieeexplore.ieee.org/abstract/document/8350274?casa_token=sAi0abbaG4UAAAAA:JpFKInOkC4lB5LJUNnH9_IBjs4ukWIzDMX3mbjgQWiRCZHweKIIkAHQPxRwkZHr6PlgZuvEPbS0f) - [Point-Based Methodology to Monitor and Control Gene Regulatory Networks via Noisy Measurements](https://ieeexplore.ieee.org/abstract/document/8265205?casa_token=K35YaaFWb2YAAAAA:jTOlvtjmUXfr8-mJugMpwQQzEqtfCLP6nxCroMgs9XSvm7j0osKiMuPr9ODPtXwrle-OIkBsArv2) - [Gene regulatory network state estimation from arbitrary correlated measurements](https://asp-eurasipjournals.springeropen.com/articles/10.1186/s13634-018-0543-y) - [Optimal Finite-Horizon Perturbation Policy for Inference of Gene Regulatory Networks](https://ieeexplore.ieee.org/abstract/document/9171420?casa_token=gXw4u7y3500AAAAA:gxm6hdOU0VzzdxuvQ460Df1_idg7SCTfSL407qNe8hv4l13jPmX_uAWFAiWm_MNzKn6M3D-pGYUR) - [State-feedback control of Partially-Observed Boolean Dynamical Systems using RNA-seq time series data](https://ieeexplore.ieee.org/abstract/document/7524920?casa_token=4MwHk1_7zHgAAAAA:lP87L56T6GF9dSbK3ZRULmNawUQtTfePon_myAydANZbtG5Bf6H7fuQ3mhDc0vDNntFg5VBmNJPHrA) - [Understanding adaptive immune system as reinforcement learning](https://journals.aps.org/prresearch/pdf/10.1103/PhysRevResearch.3.013222) - [EteRNA-RL: Using reinforcement learning to design RNA secondary structures](https://web.stanford.edu/class/cs234/CS234Win2018/past_projects/2017/2017_Kauvar_Richman_Allen_EteRNA-RL_Paper.pdf) - [Learning to Design RNA](https://arxiv.org/abs/1812.11951) - [Regulating gene expression using optimal control theory](https://ieeexplore.ieee.org/document/1188968) - [Optimal control of gene mutation in DNA replication](https://pubmed.ncbi.nlm.nih.gov/22454557/) - [Optimal control of an HIV immunology model](https://onlinelibrary.wiley.com/doi/pdf/10.1002/oca.710) - [Generative ODE Modeling with Known Unknowns, Ori Linial et al. 2020](https://arxiv.org/abs/2003.10775) c.elegans transcriptomics (the worm is much simpler than other organisms): - [Molecular topography of an entire nervous system, Taylor at al., 2021, Cell](https://www.sciencedirect.com/science/article/pii/S0092867421007583?via%3Dihub)