Interpretable CNN for ischemic stroke subtype classification with active model adaptation

Background TOAST subtype classification is important for diagnosis and research of ischemic stroke. Limited by experience of neurologist and time-consuming manual adjudication, it is a big challenge to finish TOAST classification effectively. We propose a novel active deep learning architecture to classify TOAST. Methods To simulate the diagnosis process of neurologists, we drop the valueless features by XGB algorithm and rank the remaining ones. Utilizing active learning framework, we propose a novel causal CNN, in which it combines with a mixed active selection criterion to optimize the uncertainty of samples adaptively. Meanwhile, KL-focal loss derived from the enhancement of Focal loss by KL regularization is introduced to accelerate the iterative fine-tuning of the model. Results To evaluate the proposed method, we construct a dataset which consists of totally 2310 patients. In a series of sequential experiments, we verify the effectiveness of each contribution by different evaluation metrics. Experimental results show that the proposed method achieves competitive results on each evaluation metric. In this task, the improvement of AUC is the most obvious, reaching 77.4. Conclusions We construct a backbone causal CNN to simulate the neurologist process of that could enhance the internal interpretability. The research on clinical data also indicates the potential application value of this model in stroke medicine. Future work we would consider various data types and more comprehensive patient types to achieve fully automated subtype classification.

Imprinting on time-structured acoustic stimuli in ducklings

Filial imprinting is a dedicated learning process that lacks explicit reinforcement. The phenomenon itself is narrowly heritably canalized, but its content, the representation of the parental object, reflects the circumstances of the newborn. Imprinting has recently been shown to be even more subtle and complex than previously envisaged, since ducklings and chicks are now known to select and represent for later generalization abstract conceptual properties of the objects they perceive as neonates, including movement pattern, heterogeneity and inter-component relationships of same or different. Here, we investigate day-old Mallard ( Anas platyrhynchos ) ducklings’ bias towards imprinting on acoustic stimuli made from mallards’ vocalizations as opposed to white noise, whether they imprint on the temporal structure of brief acoustic stimuli of either kind, and whether they generalize timing information across the two sounds. Our data are consistent with a strong innate preference for natural sounds, but do not reliably establish sensitivity to temporal relations. This fits with the view that imprinting includes the establishment of representations of both primary percepts and selective abstract properties of their early perceptual input, meshing together genetically transmitted prior pre-dispositions with active selection and processing of the perceptual input.

Initiating scientific collaborations across career levels and disciplines – a network analysis on behavioral data

Collaborations are essential in research, especially in answering increasingly complex questions that require integrating knowledge from different disciplines and that engage multiple stakeholders. Fostering such collaboration between newcomers and established researchers helps keep scientific communities alive while opening the way to innovation. But this is a challenge for scientific communities, especially as little is known about the onset of such collaborations. Prior social network research suggests that face-to-face interaction at scientific events as well as both network-driven selection patterns (reciprocity and transitivity) and patterns of active selection of specific others (homophily / heterophily) may be important. Learning science research implies, moreover, that selecting appropriate collaboration partners may require group awareness. In a field study at two scientific events on technology-enhanced learning (Alpine Rendez-Vous 2011 and 2013) including N = 5736 relations between 287 researchers, we investigated how researchers selected future collaboration partners, looking specifically at the role of career level, disciplinary background, and selection patterns. Face-to-face contact was measured using RFID devices. Additionally, a group awareness intervention was experimentally varied. Data was analyzed using RSiena and meta-analyses. The results showed that transitivity, reciprocity and contact duration are relevant for the identification of new potential collaboration partners. PhD students were less often chosen as new potential collaboration partners, and researchers with a background in Information Technology selected fewer new potential collaboration partners. However, group awareness support balanced this disciplinary difference. Theoretical, methodological, and practical implications of these findings are discussed.

The Sediment Selectivity of Perinereis aibuhitensis Larvae: Active or Passive?

The selectivity of Perinereis aibuhitensis larvae on different sediment types was studied using an experimental behavioral device in the lab. There were six types of sediment with different organic matter content: 2.19, 2.30, 2.86, 3.25, 3.51, and 5.52%. The results indicated significant differences in the six treatments’ organic matter content (p < 0.05). When the P. aibuhitensis larvae initially attached to the sediment, the larvae’s density showed no significant difference among the six treatments. The density of larvae decreased gradually during the experimental period. It increased with the increasing organic matter content in sediment at every sampling time, but there was no significant difference (p > 0.05). The larvae’s specific growth rate in the first month was significantly higher than those in the second and third months (p < 0.05). The mortality showed no significance at different sediments in equal sampling times, but the mortality was lower in high organic matter content sediments. This study showed that the P. aibuhitensis larvae did not make an active selection; random selection happened when initially attached to the sediment with different organic matter contents. Higher organic matter content in the sediment was more conducive to larvae survival, and the organic matter content is the limitation factor on the mortality and the density. The different densities in the natural habitat of P. aibuhitensis might occur due to the passive selection by the environment.

The mechanism of the nucleo-sugar selection by multi-subunit RNA polymerases

RNA polymerases (RNAPs) synthesize RNA from NTPs, whereas DNA polymerases synthesize DNA from 2′dNTPs. DNA polymerases select against NTPs by using steric gates to exclude the 2′OH, but RNAPs have to employ alternative selection strategies. In single-subunit RNAPs, a conserved Tyr residue discriminates against 2′dNTPs, whereas selectivity mechanisms of multi-subunit RNAPs remain hitherto unknown. Here, we show that a conserved Arg residue uses a two-pronged strategy to select against 2′dNTPs in multi-subunit RNAPs. The conserved Arg interacts with the 2′OH group to promote NTP binding, but selectively inhibits incorporation of 2′dNTPs by interacting with their 3′OH group to favor the catalytically-inert 2′-endo conformation of the deoxyribose moiety. This deformative action is an elegant example of an active selection against a substrate that is a substructure of the correct substrate. Our findings provide important insights into the evolutionary origins of biopolymers and the design of selective inhibitors of viral RNAPs.

Active Selection Constraints for Semi-supervised Clustering Algorithms

Semi.-supervised clustering algorithms aim to enhance the performance of clustering using the pairwise constraints. However, selecting these constraints randomly or improperly can minimize the performance of clustering in certain situations and with different applications. In this paper, we select the most informative constraints to improve semi-supervised clustering algorithms. We present an active selection of constraints, including active must.-link (AML) and active cannot.-link (ACL) constraints. Based on Radial-Bases Function, we compute lower-bound and upper-bound between data points to select the constraints that improve the performance. We test the proposed algorithm with the base-line methods and show that our proposed active pairwise constraints outperform other algorithms.