TIPE3��is a protein that helps regulate important cell signaling pathways involved in growth, survival, and communication. Increasing evidence suggests that it plays a significant role in the development and progression of many cancers, making it a potential target for future precision medicine approaches. This review summarizes current research on TIPE3 and its involvement in a wide range of cancers, including lung, breast, pancreatic, colorectal, ovarian, cervical, and gastric cancers, as well as glioblastoma and acute myeloid leukemia. In most cancers, TIPE3 appears to promote tumor growth, spread, resistance to treatment, changes in the immune environment, and other processes that support cancer progression. However, the authors note that TIPE3 may have the opposite effect in certain cancers, such as��head and neck squamous cell carcinoma, highlighting that its role can vary depending on the type of tumor and its biological context. The review also examines conflicting findings, particularly in colorectal cancer, and discusses possible reasons for these differences, including how TIPE3 is measured, where it is located within cells, and characteristics of the surrounding tumor environment. Finally, the authors outline future strategies for studying and targeting TIPE3, including improved diagnostic methods, advanced molecular profiling, artificial intelligence–assisted drug discovery, and combination therapies. Although no TIPE3-targeted treatments are currently available, the evidence suggests that TIPE3 has strong potential as both a biomarker and a therapeutic target in precision oncology.

Figure 1. TIPE3-mediated signaling pathways and localization-dependent functions in cancer

High-energy collisions between��protons���ǰ���heavy ions��can produce energetic��photons��(particles of light) and��jets, which are sprays of particles created when quarks or gluons are produced in a collision. Studying these��photon-triggered jetshelps physicists understand how jets lose energy as they travel through the extremely hot, dense state of matter known as the��quark-gluon plasma, which is created in heavy-ion collisions. In this study, the researchers used computer simulations based on a multistage model of jet evolution to predict several properties of photon-triggered jets and compared the results with measurements from the��ATLAS,��CMS, and��STAR��experiments. The model closely matched experimental results for several key measurements, including how often photon-triggered jets are observed, the relationship between the energies of the jet and the accompanying photon, and the angular separation between them. The researchers also found that including additional sources of photons—specifically��bremsstrahlung photons��(produced when charged particles are deflected) and photons from particle decays—improved agreement with experimental data, particularly in certain energy ranges. These findings demonstrate that the multistage model can accurately describe photon-triggered jet behavior and highlight the importance of accounting for multiple photon sources when studying jet energy loss in heavy-ion collisions.

Fig 1

Nuclear modification factor��𝑅𝐴⁢𝐴��as a function of jet-�� 𝑝𝑇��for photon-triggered jet for central��0–10%��(left) and semicentral��10–30%��(right) Pb-Pb collisions at��√𝑠𝑁⁢𝑁=5.02��TeV. The results from��matter�� +lbt��within��jetscape��for full events (solid lines) and prompt photon events (dashed lines) are compared with ATLAS data��[66].

Identifying proteins in the blood that signal early��cardiovascular disease (CVD)��could improve the ability to predict disease risk before symptoms develop. This study examined whether blood protein levels were associated with early markers of��atherosclerosis��(the buildup of plaque in the arteries) in 342 Afro-Caribbean men participating in the Tobago Health Study. The researchers measured nearly 100 cardiovascular-related proteins and compared them with three indicators of arterial health:��pulse wave velocity (PWV), which measures arterial stiffness;��coronary artery calcification (CAC), a marker of plaque in the heart’s arteries; and��abdominal aortic calcification (AAC), a marker of plaque in the body’s largest artery. After accounting for established cardiovascular risk factors such as age, blood pressure, diabetes, smoking, and cholesterol levels, they identified 18 proteins associated with arterial stiffness and one associated with coronary artery calcification. Thirteen of these protein associations had not been reported previously. Although many of the identified proteins are involved in biological processes linked to cardiovascular disease, such as inflammation, the strongest associations involved proteins not previously connected to these measures of atherosclerosis. As the first large-scale blood protein study conducted in an Afro-Caribbean population, these findings suggest that protein biomarkers of cardiovascular disease may differ across racial and ethnic groups and could help improve future approaches to risk prediction in populations with a high burden of cardiovascular disease.

FIGURE 1

Plot of protein associations with PWV in fully adjusted models. Markers labeled in blue are significant in final PWV model after Benjamini–Hochberg FDR correction. Dashed red line corresponds to the��p��value of 0.05 (−log10(0.05) ≈ 1.30).

Ashwin, T. S.; Sanda, Nihar; Coursey, Austin; Gupta, Vaibhav; Biswas, Gautam. (2026).��.��Proceedings of the ACM Symposium on Applied Computing, 87–94.��

Creating accurate��3D ultrasound��images from multiple��2D freehand ultrasound��scans is challenging because slight errors in the position or orientation of the ultrasound probe can cause the images to become misaligned, leading to distortions in the final reconstruction. Recent approaches have used��neural radiance fields (NeRFs)—an artificial intelligence technique that learns a continuous 3D representation from 2D images—but these methods are particularly sensitive to positioning errors. To address this problem, the researchers developed��GAU-NeRF (Geometric Aware Ultrasound NeRF), a new approach that stabilizes the model during training and improves its ability to correct probe position errors. The method was evaluated using both simulated and real ultrasound datasets and consistently outperformed existing reconstruction techniques. Compared with previous methods, GAU-NeRF substantially improved image quality, including increases of up to 132% in��peak signal-to-noise ratio��and 133% in the��structural similarity index, while reducing image reconstruction errors by up to 350% based on a perceptual image quality metric. These findings suggest that GAU-NeRF can produce more accurate and reliable 3D ultrasound reconstructions, which could improve applications that rely on freehand ultrasound imaging.

Figure 1.��(a) Overlap of two freehand US images acquired during two different sweep sequences for a tissue-mimicking abdominal phantom, where misregistration between two sweeps occurs. (b) Reconstruction using distance weighting [], which shows incorrect 3D geometry. (c) Reconstruction result from Ultra-NeRF. (d) Reconstruction with our GAU-NeRF method, which correctly recovers the underlying 3D structure.

Magnetic resonance imaging (MRI)��often uses��inductively coupled coils—small receiver coils placed near the area being imaged—to improve image quality. However, when these coils are positioned close to the scanner’s primary coil, they can interfere with each other, causing effects that have traditionally been viewed as reducing image quality. This study investigated why inductively coupled coils can still perform well despite this strong interaction and examined the role of modern MRI��preamplifiers��(electronic components that amplify weak signals from the coils). The researchers tested different coil configurations and preamplifier settings in laboratory experiments and validated their findings with MRI scans at 7 tesla, a high-field MRI system. They found that modern low-input-impedance preamplifiers largely prevented the signal losses typically associated with strong coil coupling, allowing the secondary coils to function effectively even when placed very close to the primary coil. Although the interaction between the coils altered the electrical properties of the primary coil, it had little effect on the overall��signal-to-noise ratio (SNR), a key measure of image quality. In contrast, reducing the effectiveness of the preamplifiers led to a 21%–23% decrease in SNR. These findings suggest that modern preamplifiers play a critical role in maintaining MRI performance and could simplify the design of inductively coupled coils for future imaging systems.

FIGURE 1

(A) Setup and results of measuring the impedance of a 10-cm-diameter circular 7 T RF coil on a bottle phantom. (B) Setup and results of the same coil (primary coil) when a smaller 5-cm-diameter inductively coupled coil was placed underneath the primary coil but above the phantom. The primary coil was not retuned or rematched after introducing the inductively coupled coil. (C) Simplified equivalent circuit model of the coupled inductively coupled and primary coils illustrating resonance splitting due to strong mutual coupling.

Hybrid��convolutional neural network (CNN)�����Ի���Transformer��models have become popular for analyzing 3D medical images because they combine the strengths of both approaches: CNNs are effective at identifying local image features, while Transformers capture broader patterns across an image. This study introduces��DSR (Decoder Supervision and Reconstruction), a new training strategy designed to improve how these models learn to reconstruct detailed 3D segmentations without changing how they operate once deployed. During training, DSR adds temporary components that provide extra supervision at multiple stages of the model and encourages the final image representation to remain closely aligned with the original scan. These additional components are removed before the model is used, meaning the deployed model remains the same size and complexity. The approach was evaluated on three medical imaging tasks involving abdominal��CT��scans,��MRI��scans of��vestibular schwannomas��(noncancerous tumors of the nerve connecting the ear and brain), and prostate MRI. Across all three tasks, DSR achieved the highest��Dice scores��(a measure of how closely the predicted segmentation matches the true anatomy) compared with the methods evaluated and also improved measures of boundary accuracy. These findings suggest that strengthening the training of the decoder—the part of the model that reconstructs the final segmented image—can improve the accuracy of 3D medical image segmentation without increasing the computational demands of the final model.

Fig 1

Overview of DSR-Net. The input volume is encoded by parallel 3D CNN and Swin Transformer encoders, whose multi-scale features are fused and passed to a shared convolutional decoder. In the decoder,����initializes the first decoder block, while��,��,��, and����are fused into the subsequent stages that produce��,��,��, and��, respectively. DSR adds training-time decoder branches: auxiliary segmentation heads supervise intermediate decoder features, and an image reconstruction head regularizes the highest-resolution decoder feature. These auxiliary branches are removed at inference, leaving only the main encoder, decoder, and final segmentation head

Memory decline is one of the hallmark features of��Alzheimer’s disease (AD), but measurable changes in memory often occur only after significant changes have already taken place in the brain. This study investigated whether brain characteristics seen in��midlife��share genetic links with memory performance later in life, with the goal of identifying earlier markers of Alzheimer’s-related cognitive decline. The researchers analyzed genetic data from more than 24,000 older adults alongside brain imaging and genetic data from over 33,000 middle-aged participants in the UK Biobank. They found that brain features related to the structure and microscopic organization of the��medial temporal lobe—a region critical for memory—and the��frontal lobe��showed the strongest shared genetic links with memory performance. They also identified shared genetic patterns involving the��default mode network, a network of brain regions that is important for memory and is known to be affected early in Alzheimer’s disease. These findings suggest that changes in specific brain regions during midlife may reflect genetic pathways that contribute to later-life memory decline and Alzheimer’s disease, potentially helping researchers identify earlier biomarkers and new targets for treatment.

FIGURE 1

Genetic covariance between imaging-derived phenotypes (IDPs) and memory performance. Volcano plots show strength of genetic covariance of diffusion (top panel), structural (middle panel), and functional (bottom panel) IDPs with cross-sectional memory performance (MEM) for all, impaired, and unimpaired individuals, including the��APOEregion in the analyses. Colors highlight significant genetic covariance with memory (FDR-corrected��p��<��0.05), with green indicating positive covariance and red indicating negative covariance. Data point shape for structural IDPs indicates the atlas used to generate the cortical measure, with circles representing the Desikan-Killiany atlas, squares representing the Destrieux atlas, diamonds representing the Desikan-Killiany-Tourville (DKT) atlas, and triangles representing other smaller atlases (e.g., thalamic nuclei, hippocampal subfields). Data point shape for functional IDPs indicates the dimensionality of group ICA, with diamonds representing 100-dimensional ICA and circles representing 25-dimensional ICA. Top IDPs for each modality and analysis that survived FDR correction are labeled. Diffusion measures highlight regions such as the fornix and corona radiata; structural measures highlight regions such as the lingual gyrus, anterior cingulate gyrus and sulcus, fusiform gyrus, and superior temporal gyrus and sulcus. Table����provides the mapping between original UKB IDP names and our intuitive names. Ant., anterior; L, left hemisphere; Lat., lateral; Post., posterior; R, right hemisphere; Sup., superior.

Some people experience serious side effects from��azathioprine, a medication commonly used to treat autoimmune diseases, and genetic differences can influence that risk. While variants in the��TPMT�����Ի���NUDT15��genes are already known to increase the likelihood of��myelotoxicity��(damage to the bone marrow that can reduce blood cell production), other genetic factors remain less well understood. In this study, the researchers used��PrediXcan, a computational method that estimates how strongly a person’s genes are likely to be expressed based on their genetic data, to investigate whether predicted gene expression was associated with known azathioprine side effects. They analyzed data from 1,364 people who had recently started azathioprine and found that individuals with the lowest predicted expression of the��TPMT��gene had more than three times the odds of developing��leukopenia��(a low white blood cell count) compared with those with the highest predicted TPMT expression. No significant associations were found for other known side effects. These findings suggest that approaches such as PrediXcan may help identify additional genetic factors that influence how people respond to medications, supporting more personalized treatment in the future.

Fig 1

Predicted expression of��TPMT��in liver tissue by��TPMT��phenotype group

]]> /valiant/2026/07/28/beyond-the-response-examining-reasoning-and-execution-fidelity-in-large-language-models-for-mental-health/ Tue, 28 Jul 2026 19:52:50 +0000 /valiant/?p=7218 Qadir, Sarvech; Ni, Congning; Vaidya, Mihir Sachin; Ryu, Hyeyoung; Mulvaney, Shelagh A.; Kantarcioglu, Murat; Novak, Laurie Lovett; Malin, Bradley; Rose, Susannah Leigh; Yin, Zhijun. (2026).��.��Proceedings of the 9th ACM Conference on Fairness, Accountability, and Transparency (FAccT 2026), 6702–6722.��

As��large language models (LLMs)��are increasingly used to provide mental health support, ensuring that they respond safely and consistently has become an important concern. Most evaluations focus on the quality of a model’s final answer, but this study examined whether the models actually followed the plans they appeared to make while generating those responses. The researchers introduced the concept of��execution fidelity, which measures how well a model’s final response aligns with the commitments it expressed during its internal reasoning process, such as showing empathy, offering guidance, setting appropriate boundaries, or framing a problem. Five leading LLMs were evaluated using 75 high-risk mental health scenarios involving depression, anxiety, trauma, and emotional distress. The analysis found that models often differed in how they planned their responses and, in some cases, failed to carry through on important safety-related commitments. Common issues included omitting disclaimers, providing less detailed guidance than originally planned, or replacing specific recommendations with more general emotional reassurance. The researchers also found that greater uncertainty and inconsistency in a model’s planning were associated with a higher likelihood of these gaps. These findings suggest that examining how AI systems develop their responses—not just the final output—may help identify safety risks and improve oversight of LLMs used in mental health settings.