130,909 research outputs found
Optimization of RPMI 2650 Cells as a Model for Nasal Mucosa
No full textIn the past few years, a human nasal epithelial cell line derived from septum carcinoma (RPMI 2650) has been proposed as a potential in vitro model for screening nasally delivered drugs. However, these studies have left some unanswered questions in terms of the validation of the in vitro model.
In particular, no clear agreement was found with respect to several parameters, such as the seeding density, the time for switching cell culture from liquid covered culture (LCC) to air liquid interface (ALI) conditions, or the day at which cell cultures have to be used for transport experiments, when these cells are cultured in vitro.
Hence, the aim of this study was to expand on the previous in vitro cell models to better define the fundamental parameters to be used as a tool for studying drug deposition and transport through the nasal mucosa
La struttura assente e il principio di immanenza. Qualche riflessione sul metodo semiotico
No full textAnalytic semiotics has always created structures for describing and analysing texts. As was inevitable, ontological questions have emerged on these structures' modes of existence: are they real or abstract? Do they exist in texts, or inside people's heads? Umberto Eco reflected on the status and location of structures almost fifty years ago in his book La struttura assente, in a section of the book that became famous, Section D, which is dedicated to the epistemology of structural models (Eco 1968). In this article I take as my starting point some observations made by Eco at the end of the 1960s in order to see how some assumptions of structuralism influenced the development of semiotics over the decades that followed
Efficient Token Pruning in Vision Transformers Using an Attention-Based Multilayer Network
No full textVision Transformers (ViTs), although very successful, have a major limitation to overcome, namely the need for significant computational resources to use them. Several approaches have been proposed to limit the resources required to work with ViTs, aiming at pruning the data provided in input to them. In this paper, we propose Token Reduction via an Attention-based Multilayer network (TRAM), the first approach that achieves this goal using a multilayer network-based representation of the attention matrices. TRAM can work with most ViTs without the need for fine-tuning. It makes several contributions to the literature in this research area; in particular, it is characterized by: (i) a new representation of ViTs based on a multilayer network; (ii) a new approach to evaluate the relevance of tokens based on a new centrality measure computed on the multilayer network; and (iii) an approach to reduce the number of tokens based on this centrality measure. We have validated TRAM by comparing it with several state-of-the-art approaches during an extensive experimental campaign carried out on different image datasets. The results obtained demonstrate not only the efficiency but also the effectiveness of TRAM in reducing the computational load of ViTs while still allowing them to provide accurate result
Speeding up Vision Transformers Through Reinforcement Learning
No full textIn recent years, Transformers have led a revolution in Natural Language Processing, and Vision Transformers (ViTs) promise to do the same in Computer Vision. The main obstacle to the widespread use of ViTs is their computational cost. Indeed, given an image divided into a list of patches, ViTs compute, for each layer, the attention of each patch with respect to all others. In the literature, many solutions try to reduce the computational cost of attention layers using quantization, knowledge distillation, and input perturbation. In this paper, we aim to make a contribution in this setting. In particular, we propose AgentViT, a framework that uses Reinforcement Learning to train an agent whose task is to identify the least important patches during the training of a ViT. Once such patches are identified, AgentViT removes them, thus reducing the number of patches processed by the ViT. Our goal is to reduce the training time of the ViT while maintaining competitive performanc
Integrating Gradient and Mask-based Approaches for Vision Transformer Explainability
No full textVision Transformers (ViTs) have demonstrated outstanding performance across different computer vision tasks thanks to their self-attention mechanism that captures long-range dependencies effectively. However, the inherent complexity of ViTs presents significant challenges in explaining their outputs, which is fundamental in safety-critical domains. To tackle the challenge of explaining ViT outputs, this paper presents GradMask, a novel method that integrates gradients into the mask generation process to create explanation heatmaps. GradMask uses the query, key, and value matrices from each attention layer and computes their gradients with respect to a target class. Afterward, it uses these gradients to generate binary masks, which are then weighted by the corresponding ViT's confidence scores. Finally, it combines the weighted masks to generate the resulting heatmap. Experimental evaluations on an ImageNet subset with ViT and DeiT (Data-efficient Image Transformer) architectures show that GradMask achieves competitive performance according to standard explainability metrics, such as Insertion, Deletion, and Pointing Game. A hyperparameter analysis confirms the high computational efficiency of GradMask, while an ablation study highlights the importance of combining gradients and masks for the generation of the explanation heatmap. Finally, a qualitative analysis shows the improved explainability of GradMask compared to existing methods, making it a promising approach for understanding ViTs
Speeding up Vision Transformers Through Reinforcement Learning
No full textIn recent years, Transformers have led a revolution in Natural Language Processing, and Vision Transformers (ViTs) promise to do the same in Computer Vision. The main obstacle to the widespread use of ViTs is their computational cost. Indeed, given an image divided into a list of patches, ViTs compute, for each layer, the attention of each patch with respect to all others. In the literature, many solutions try to reduce the computational cost of attention layers using quantization, knowledge distillation, and input perturbation. In this paper, we aim to make a contribution in this setting. In particular, we propose AgentViT, a framework that uses Reinforcement Learning to train an agent whose task is to identify the least important patches during the training of a ViT. Once such patches are identified, AgentViT removes them, thus reducing the number of patches processed by the ViT. Our goal is to reduce the training time of the ViT while maintaining competitive performance
Adaptive Patch Selection to Improve Vision Transformers through Reinforcement Learning
No full textIn recent years, Transformers have revolutionized the management of Natural Language Processing tasks, and Vision Transformers (ViTs) promise to do the same for Computer Vision ones. However, the adoption of ViTs is hampered by their computational cost. Indeed, given an image divided into patches, it is necessary to compute for each layer the attention of each patch with respect to all the others. Researchers have proposed many solutions to reduce the computational cost of attention layers by adopting techniques such as quantization, knowledge distillation and manipulation of input images. In this paper, we aim to contribute to the solution of this problem. In particular, we propose a new framework, called AgentViT, which uses Reinforcement Learning to train an agent that selects the most important patches to improve the learning of a ViT. The goal of AgentViT is to reduce the number of patches processed by a ViT, and thus its computational load, while still maintaining competitive performance. We tested AgentViT on CIFAR10, FashionMNIST, and Imagenette+ (which is a subset of ImageNet) in the image classification task and obtained promising performance when compared to baseline ViTs and other related approaches available in the literatur
Biological and molecular characterization of in silico identified putative inhibitors of paraspeckle assembly with potential anti-multiple myeloma activity.
No full textIntroduction
Multiple myeloma (MM) is the second most common haematological malignancy, being characterized by abnormal proliferation of plasma cells (PCs) predominantly within the bone marrow.
A new cellular organelle named paraspeckle (PS) has been recently associated with several biological processes, including DNA damage systems regulation and stress response. The assembly of these nuclear bodies requires the presence of the essential long noncoding RNA (lncRNA) NEAT1 and seven PS proteins (PSPs), among which NONO and SFPQ.
The relevance of PSs in MM pathogenesis has been well documented. Indeed, the crucial scaffold NEAT1 has been found to be significantly overexpressed in PCs of MM patients with respect to the normal counterpart and its silencing has been shown to affect malignant PCs proliferation and viability, triggering anti-tumour activity, both in vitro and in vivo. On the contrary, NEAT1 transactivation induced by stressful conditions has been associated with increased PSPs levels and enhanced MM cells viability, suggesting a pro-survival and anti-apoptotic role of PSs in MM.
Similarly, we recently reported higher NONO expression in primary CD138+ MM cells, correlating with poorer OS and PFS.
Aim
Because of the genetic complexity of MM cells, it is still difficult to imagine a universal therapeutic approach able to effectively target all the malignant subclones. As a result, patients inevitably relapse and become resistant to subsequent lines of therapies, highlighting the urgent clinical need to identify novel druggable vulnerabilities. Our hypothesis is that PSs could represent a novel specific vulnerability in MM across genetic subgroups. However, clinical translation of lncRNA targeting may be problematic. On the contrary, protein components of PSs are more amenable to drug design. Therefore, we set out to perform an in silico screening of small molecules targeting crucial interaction points between NEAT1, NONO, and SFPQ and supposed to disrupt PSs structure, thus, at least in part, mimicking the NEAT1 silencing strategy.
The aim of this study is to assess in vitro the on-target activity of the compounds representing the top 6 hits of the screening.
Materials and Methods
The biological activity of the 6 small molecules was evaluated in a panel of 6 MM cell lines (HMCLs), 4 haematological non-HMCLs and 6 healthy donors-derived PBMC samples. Dose-effect curves for the determination of IC50 values were obtained by Trypan Blue exclusion cell counts and CellTiter-Glo assay. Alteration of PSs integrity was evaluated by NEAT1 RNA-FISH and NONO IF by confocal microscopy. Clonogenic potential was evaluated through methylcellulose assay. Cell cycle phases modulation and apoptosis induction were investigated by FACS analysis. The levels of PSPs were assessed by means of WB analysis and IF technique. Transcriptome analysis was performed by means of ClariomTM D arrays on Affymetrix platform.
Results
2 of the 6 compounds demonstrated a significant biological activity after 5 days of treatment in all the tested HMCLs but not in non-MM cells and healthy donors-derived PBMCs (≈75% vs. 25% of affected cellular fraction, respectively), confirming a specific anti-MM effect.
Confocal microscopy experiments showed a ≈45% decrease in the number of PSs per cell in treated HMCLs, validating the expected loss of PS integrity and the on-target activity of both molecules.
In line with biological data, cell cycle analysis revealed a perturbation of the cell distribution upon small compounds treatment of HMCLs. Indeed, HMCLs displayed a significant increase of the cellular population distributed in the Sub-G0/G1 phase, and a downregulation of the percentage of cells in the S phase. FACS data were in agreement with clonogenic potential results showing a median of 50 colonies in vehicle-treated samples vs. 9 and 13 upon treatment with the 2 inhibitors.
Flow cytometry analysis revealed a dose-dependent increase of apoptotic cells in treated HMCLs (2/5-fold increase, depending on the HMCL tested), suggesting a pro-apoptotic effect of both inhibitors on HMCLs.
Additionally, we highlighted a decrease of key and core-localizing PSPs levels after treatment, among which NONO and SFPQ.
Remarkably, in line with previous findings highlighting a MM-specific effect of NEAT1 in a panel of HMCLs, none of the above-mentioned results were observed in non-MM cells, confirming the specificity of both inhibitors for HMCLs.
Finally, preliminary transcriptomic analysis revealed the downregulation of several pathways associated with tumorigenic processes.
Discussion and Conclusions
In the present study, we assessed the biological and molecular activity of the top 6 candidates resulted from an in silico screening of commercially available drug-like small molecules potentially affecting PSs structure integrity. We identified 2 molecules that, by specifically targeting NEAT1 essential protein interactors, exert an anti-MM specific activity resulting in PS structural core impairment and apoptosis induction. These promising preliminary results emphasize the need of a better comprehension of the mechanisms underlying the activity of the novel identified small compounds and suggest PS targeting could represent a possible new druggable vulnerability in MM. As a matter of fact, PSs targeting may have a great translational relevance since cytogenetically different cell lines resulted to be responsive to the selected inhibitors. Additionally, stressful conditions promoting PSs assembly are often associated with more aggressive tumor stages and chemoresistance mechanisms, raising the possibility to use this innovative therapeutic strategy in all the subsets of MM patients
Are Large Language Models Better Peer-Reviewers Than Humans? An Early Investigation on OpenReview
No full textIn recent years, Large Language Models (LLMs) have often been used by paper reviewers, despite this practice being generally prohibited. This has raised, and continues to raise, issues concerning ethics, review reliability, and the risk of review manipulation. Indeed, several arXiv preprints were recently discovered to contain invisible, LLM-targeted instructions designed to persuade an AI reviewer to yield a positive review. In this paper, we propose a systematic analysis of LLMs’ review capabilities in this complex and evolving scenario. In particular, we want to address two research questions: (i) How can LLM ratings be compared with human ratings?, and (ii) Can hidden positive prompts injected in a manuscript alter an LLM’s generated review? To address these questions, we created a dataset of 400 papers from OpenReview. For each paper, this dataset contains human reviews and scores already present in OpenReview, as well as reviews performed by three state-of-the-art LLMs, added by us. Our results show that human reviewers assign higher and more widely dispersed scores that clearly distinguish accepted and rejected papers. In contrast, LLM ratings cluster close to their mean value, blurring the distinction between accepted and rejected papers. Furthermore, a negative prompt given by the reviewer makes the LLM lower its scores, while a hidden positive prompt injected by the author often fails to raise scores, and sometimes triggers even lower scores, if detected by the LLM. These results reveal both the potential and fragility of delegating peer review tasks to LLMs
- …
