Search

The data explosion and development of artificial intelligence (AI) has fueled the demand for recommendation systems, information retrieval, personalization, among others. Consequently, the need of a solution to optimize these systems “on-the-fly” has also grown rapidly. Contextual bandit is a machine learning framework designed to tackle complex situations in an online manner, where the agent can select actions (i.e., arms) based on available context information. Based the feedback, the agent can learn the relations between context information and rewards for each arm, which further improves arm selection in the future. In practice, however, the learning environment may be far from being perfect. For example, the available context information may not be accurate, the reward feedback may be delayed or even missing, and data may not be centrally available due to user privacy concerns.In this dissertation, we consider the practical scenario of contextual bandits in an imperfect environment. First, we focus on imperfect context and study learning with probabilistic contexts, where a bundle of contexts are revealed to the agent along with their corresponding probabilities instead of true context. Second, we study reward imperfectness by considering delayed or missing reward feedback. Third, we turn to an adversarial environment and study a novel combinatorial setting with arm removal and submodular utility where some selected arms can be removed adversarially. Finally, we consider a privacy-preserving federated bandit where a group of agents cooperate to solve the bandit problem, while ensuring that their communication remains private. For each of the settings, we propose new learning algorithms, analyze the cumulative regret, and conduct empirical evaluations based on real-world applications.

Luting Yang, Yaping Yan Key Laboratory of the Ministry of Education for Medicinal Resources and Natural Pharmaceutical Chemistry, National Engineering Laboratory for Resource Development of Endangered Crude Drugs in Northwest of China, College of Life Sciences, Shaanxi Normal University, Xi’an, People’s Republic of ChinaCorrespondence: Luting Yang; Yaping Yan, Email yangluting@snnu.edu.cn; yaping.yan@snnu.edu.cnAbstract: Post-translational modifications (PTMs) of proteins represent as a key step in regulating their biological functions and dynamic interaction with other players. This process is fine-tuned by a myriad of enzymes named “writers, readers and erasers” whose actions are precisely controlled. Either the mutation, aberration in the expression of the aforementioned enzymes or their substrates have shown to participate in the pathogenesis of various skin diseases such as melanoma, vitiligo, psoriasis, eczema, atopic dermatitis and inherited dermatological diseases. It is becoming increasingly clear that key transcriptional factors, inflammation-related molecules are prone to PTMs. Despite their importance in regulating key processes including inflammation, keratinocyte apoptosis, proliferation and differentiation, PTMs have received less attention due to the challenges involved. Here in this review we summarize the role of the most common types and the newly discovered PTMs, including acetylation, glycosylation, citrullination, PARylation and sumoylation in dermatoses and surveys the recent progress in PTM-based therapeutic approaches in skin diseases.Keywords: post-translational modification, skin diseases, molecular mechanism, therapeutics

Keratin 17 (K17) is a type I Intermediate filament which is highly expressed in psoriatic epidermis. Studies on K17 uncover several role for K17, including mechanical support, regulation of cell proliferation and differentiation, and apoptosis. We previously proposed a K17/T-cell/cytokine autoimmune loop in which K17 could be induced by inflammatory cytokine stimulation in psoriasis. Yet, little is known about the regulation of K17 expression via post-translational modification. Ubiquitination is one of the most important post-translational modifications that modulate the functions of proteins. UbiNet software predicated several ubiquitination sites on human K17. We sought to determine whether K17 underwent ubiquitination by dysregulated E3 ligase in psoriasis, which regulated the protein levels of K17 and contributed to the pathogenensis of psoriasis. To assess whether K17 is ubiquitinated in human keratinocytes, we employed human keratinocyte cell line HaCaT cells. Furthermore, we tested K17 ubiquitination levels in normal human keratinocyte and psoriatic keratinocytes by immunoprecipitation. Mass spectrometry was used to identify the E3 ligases of K17. We initially found that K17 is ubiquitinated in human keratinocytes, while with a higher level of ubiquitination observed in psoriatic keratinocytes. Moreover, immunoprecipitation coupled with mass spectrometry identified the candidate E3 ligase for K17, Trim21. We further confirmed the direct interaction between Trim21 and K17 by immunoprecipitation and immunofluorescence in HaCaT cells. Importantly, Trim21 mRNA and protein levels were significantly elevated in psoriatic epidermis comparing with normal epidermis, as assessed by RT-PCR, western blotting, immunofluorescence and immunohistochemestry. Overexpression of Trim21 led to activation of K17, but not degradation, which can be partly illustrated by our finding that K17 is ubiquitinated by Trim21 via K63-linkage. Taken together, our data suggest a previously unrecognized role for ubiquitination in regulation protein level of K17 in psoriasis, further indicating a novel mechanism in the pathogenesis of psoriasis. Trim21, which regulated K17 activation by ubiquitination, provides a potential target for interrupting K17 activation in the treatment of psoriasis.