To obtain reliable dynamical models of intracellular gene regulation directly from expression data, one, to get the structure of the model, needs to first infer existing causal influences among the genes of interest, i.e. solve the network inference problem. Network inference is thus an enabling technology and key problem in Systems biology. Through systematic perturbation experiments, utilizing e.g. siRNA knock-downs, and measurement of the resulting gene expression changes, using e.g. qRT-PCR, followed by network inference, one could in theory test thousands of genetic interaction hypotheses in one set of experiments. Network inference could thus speed up the production of biological knowledge by several orders of magnitude.

Current expression data sets, however, suffer from (i) few data points compared to the high number of genes and possible interactions, i.e. high dimensionality but few samples, (ii) large measurement uncertainty both in the perturbations and responses, i.e. low signal to noise ratio and errors-in-variables, and (iii) redundant nearly collinear variables, i.e. ill-conditioned response matrices. The network inference methods that have been applied were not designed for data with these properties. Recent scrutiny of promising studies, such as the SOS pathway in E. coli by Gardner et al. in Science (2003) and Snf1 pathway in S. cerevisiae by Lorenz et al. in PNAS (2009), has shown that the inferred models are unreliable. Out of 35 inference methods benchmarked in the DREAM challenge only some yielded networks more similar to the true network than expected when picking links randomly.

My long-term goal is to both demonstrate, based on perturbation experiments in a human cell line and measurements of gene expression, that thousands of genetic interaction hypotheses can be tested with confidence in vitro and that reliable dynamical models can be constructed directly from data. I have already worked for a decade towards this goal and e.g. created a new mathematical theory for robust network inference, which enables testing of interaction hypotheses with confidence. The main aims of this project is to (i) develop a method for finding the most likely network based on partly informative data, (ii) find the best inference method for a variety of network and data properties, (iii) explore the limits for what can be inferred from different types of experiments, and (v) apply these methods to qRT-PCR data from in vitro experiments in epidermoid squamous cell carcinoma cell line A431.

In conclusion, I intend to realize the potential of network inference to increase generation of knowledge of genetic interactions by several orders of magnitude, which would be disruptive in Biology and Medicine.

Keywords: Systems biology, Bioinformatics, Network inference, Reverse engineering, Gene regulatory networks, Robust network inference, System identification

系統生物學中顛覆性的創新研究 – 基因間交互作用的映射及動態模組的建構

為了從實驗中得到的表達數據獲得可靠的基因調控動力模組，研究者首先需要推斷出基因間已知的影響因子來獲得模組結構，即解決網絡推理的問題。網絡推理是系統生物學中的有利技術然而是個待解決的關鍵習題。通過系統擾動試驗以及由此產生的基因表達變化的測量，隨後通過網絡推理，人們理論上可以在實驗中檢定大量基因間的相互作用並以數量級速度加快生物學知識。

現今的實驗表達數據，面臨以下問題：(一)少量樣品，但包含如基因間的交互作用等如此龐大的數據;（二）測量結果及分析的不確定性; （三）數據中含近線性相關的變量，即病態矩陣。現今應用於網絡推理方法沒有為具有這些特質的數據做特製化。近期研究已證實Gardner等學者推斷大腸桿菌的SOS途徑(Science, 2003）和Lorenz 等學者推斷S. cerevisiae的Snf1途徑 (PNAS, 2009），其推理模型是不可靠的。再者，有35種推斷方法參與DREAM challenge競賽，只有部分產生的網絡是較靠近真實網絡的，其餘的網絡，甚至劣於隨機選取連結。

為了解出網絡推理的習題，我的研究長期目標是欲證實在人類細胞株的擾動實驗和基因表達變化的測量上，能夠在體外高度信賴地檢定大量（thousands）基因間的交互作用的假說，及可以直接從資料中建構可靠的動力模組。為了實現這一目標，我已致力在此研究十年時間，並已從數學方法來建立穩健的網絡推理。此提出的計畫書將持續實現我的研究目標: (ㄧ）發展一個能夠只有部分信息數據中尋找最有可能的網絡的方法，（二）找到適用於各種網絡和數據屬性最好的推理方法，（三）探討從不同類型的實驗推斷的可能限制，以及（五）最後將此方法應用在從體外實驗表皮鱗癌細胞株A431的qRT-PCR數據。

在此提出的研究計畫書中，我將發揮網絡推理的潛力來映像(mapping)基因交互作用及建構動態模組，這將在生物學和醫學中顛覆性的創新研究。

關鍵字：系統生物學、生物資訊學、網絡推理、反向工程、基因調控網絡、穩健網絡推理、系統辨識

Principal Investigator: Ass. Prof. Torbjörn Nordling. 

Sponsor: Ministry of Science and Technology in Taiwan 

Grant identifier: 105-2218-E-006-016-MY2.

Duration: 2016-03-01 to 2018-02-28.

Collaborators: Prof. Erik Sonnhammer, Stockholm University and Ass. Prof. Mika Gustafsson, Linköping University

Members: Lewis Hsu, Tim Hsiu, Rain Wu, Eric Lee, Chi-Ching Hsu, Torbjörn Nordling.