DxTerity (United States)

Categorical Distributional Reinforcement Learning (CDRL) has demonstrated superior sample efficiency in learning complex tasks compared to conventional Reinforcement Learning (RL) approaches. However, the practical application of CDRL is encumbered by challenging projection steps, detailed parameter tuning, and domain knowledge. This paper addresses these challenges by introducing a pioneering Continuous Distributional Model-Free RL algorithm tailored for continuous action spaces. The proposed algorithm simplifies the implementation of distributional RL, adopting an actor-critic architecture wherein the critic outputs a continuous probability distribution. Additionally, we propose an ensemble of multiple critics fused through a Kalman fusion mechanism to mitigate overestimation bias. Through a series of experiments, we validate that our proposed method provides a sample-efficient solution for executing complex continuous-control tasks.

The possibility of a large-scale acute radiation exposure necessitates the development of new methods that could provide rapid individual dose estimates with high sample throughput. The focus of the study was an intercomparison of laboratories' dose-assessment performances using gene expression assays. Lithium-heparinized whole blood from one healthy donor was irradiated (240 kVp, 1 Gy/min) immediately after venipuncture at approximately 37°C using single X-ray doses. Blood samples to establish calibration curves (0.25-4 Gy) as well as 10 blinded test samples (0.1-6.4 Gy) were incubated for 24 h at 37°C supplemented with an equal volume of medium and 10% fetal calf serum. For quantitative reverse transcription polymerase chain reaction (qRT-PCR), samples were lysed, stored at -20°C and shipped on ice. For the Chemical Ligation Dependent Probe Amplification methodology (CLPA), aliquots were incubated in 2 ml CLPA reaction buffer (DxTerity), mixed and shipped at room temperature. Assays were run in each laboratory according to locally established protocols. The mean absolute difference (MAD) of estimated doses relative to the true doses (in Gy) was calculated. We also merged doses into binary categories reflecting aspects of clinical/diagnostic relevance and examined accuracy, sensitivity and specificity. The earliest reported time on dose estimates was <8 h. The standard deviation of technical replicate measurements in 75% of all measurements was below 11%. MAD values of 0.3-0.5 Gy and 0.8-1.3 Gy divided the laboratories contributions into two groups. These fourfold differences in accuracy could be primarily explained by unexpected variances of the housekeeping gene (P = 0.0008) and performance differences in processing of calibration and blinded test samples by half of the contributing laboratories. Reported gene expression dose estimates aggregated into binary categories in general showed an accuracies and sensitivities of 93-100% and 76-100% for the groups, with low MAD and high MAD, respectively. In conclusion, gene expression-based dose estimates were reported quickly, and for laboratories with MAD between 0.3-0.5 Gy binary dose categories of clinical significance could be discriminated with an accuracy and sensitivity comparable to established cytogenetic assays.

Abstract Formalin-Fixed Paraffin Embedded (FFPE) tissue samples are the standard for clinical oncology tissue sample archiving worldwide. It is estimated that there are approximately a billion FFPE tissue cancer samples archived in hospitals and tissue banks around the world, and many of these samples have full clinical annotation including long term follow-up data. Current methods of analyzing gene expression in these samples involve laborious and expensive sample preparation protocols and long turnaround times (2 days or longer). We have developed a novel, rapid (3 hours), sample-to-answer technology to analyze gene expression from clinical FFPE tissues without the need for RNA extraction. The technique is highly sensitive and requires only 2 mm x 5 mm x 10 microns of FFPE material scraped from a pathology slide. The gene expression data obtained directly from FFPE without sample preparation correlates highly with data obtained on RNA extracted from the same FFPE sample using conventional methods. Importantly, the gene expression data obtained from prostate and breast FFPE samples correlate highly with microarray data, and with data from matching mirrored FF (Fresh Frozen) samples. The assay can be developed for detection using Luminex instrument or Capillary Electrophoresis instrument (Caliper GX or ABI 3500). Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research; 2012 Mar 31-Apr 4; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2012;72(8 Suppl):Abstract nr 3204. doi:1538-7445.AM2012-3204