A major bottleneck in developing clinically impactful machine learning models is a lack of labeled training data for model supervision. Thus, medical researchers increasingly turn to weaker, noisier sources of supervision, such as leveraging extractions from unstructured text reports to supervise image classification. A key challenge in weak supervision is combining sources of information that may differ in quality and have correlated errors. Recently, a statistical theory of weak supervision called data programming has shown promise in addressing this challenge. Data programming now underpins many deployed machine-learning systems in the technology industry, even for critical applications. We propose a new technique for applying data programming to the problem of cross-modal weak supervision in medicine, wherein weak labels derived from an auxiliary modality (e.g., text) are used to train models over a different target modality (e.g., images). We evaluate our approach on diverse clinical tasks via direct comparison to institution-scale, hand-labeled datasets. We find that our supervision technique increases model performance by up to 6 points area under the receiver operating characteristic curve (ROC-AUC) over baseline methods by improving both coverage and quality of the weak labels. Our approach yields models that on average perform within 1.75 points ROC-AUC of those supervised with physician-years of hand labeling and outperform those supervised with physician-months of hand labeling by 10.25 points ROC-AUC, while using only person-days of developer time and clinician work-a time saving of 96%. Our results suggest that modern weak supervision techniques such as data programming may enable more rapid development and deployment of clinically useful machine-learning models.

Objective

Since arterial stiffness is a functional measure of arterial compliance and may be an important marker of cerebrovascular disease, we examined the association of carotid artery stiffness with white matter hyperintensity volume (WMHV) in a cross-sectional study of 1,166 stroke-free participants.

Methods

Carotid beta stiffness index (STIFF) was assessed by M-mode ultrasound of the common carotid artery and calculated as the ratio of natural log of the difference between systolic and diastolic blood pressure over STRAIN, a ratio of the difference between carotid systolic and diastolic diameter (DD) divided by DD. WMHV was measured by fluid-attenuated inversion recovery MRI. The associations of STIFF, DD, and STRAIN with WMHV were examined using linear regression after adjusting for sociodemographic, lifestyle, and vascular risk factors.

Results

In a fully adjusted model, larger carotid DD was significantly associated with greater log-WMHV (β = 0.09, p = 0.001). STIFF and STRAIN were not significantly associated with WMHV. In adjusted analyses stratified by race-ethnicity, STRAIN (β = -1.78, p = 0.002) and DD (β = 0.11, p = 0.001) were both associated with greater log-WMHV among Hispanic participants, but not among black or white participants.

Conclusions

Large carotid artery diameters are associated with greater burden of white matter hyperintensity (WMH) in this multiethnic population. The association between increased diameters, decreased STRAIN, and greater WMH burden is more pronounced among Hispanics. These associations suggest a potential important pathophysiologic role of extracranial large artery remodeling in the burden of WMH.

N-glycosylation has been shown to affect the pharmacokinetic properties of several classes of biologics, including monoclonal antibodies, blood factors, and lysosomal enzymes. In the last two decades, N-glycan engineering has been employed to achieve a N-glycosylation profile that is either more consistent or aligned with a specific improved activity (i.e., effector function or serum half-life). In particular, attention has focused on engineering processes in vivo or in vitro to alter the structure of the N-glycosylation of the Fc region of anti-cancer monoclonal antibodies in order to increase antibody-dependent cell-mediated cytotoxicity (ADCC). Here, we applied the mannosidase I inhibitor kifunensine to the Nicotiana benthamiana transient expression platform to produce an afucosylated anti-CD20 antibody (rituximab). We determined the optimal concentration of kifunensine used in the infiltration solution, 0.375 µM, which was sufficient to produce exclusively oligomannose glycoforms, at a concentration 14 times lower than previously published levels. The resulting afucosylated rituximab revealed a 14-fold increase in ADCC activity targeting the lymphoma cell line Wil2-S when compared with rituximab produced in the absence of kifunensine. When applied to the cost-effective and scalable N. benthamiana transient expression platform, the use of kifunensine allows simple in-process glycan engineering without the need for transgenic hosts.