Ennis, R. | bioRxiv Science

bioRxiv2019

The color appearance of three-dimensional, curved, transparent objects

Studies on the perceived color of transparent objects have elucidated potential mechanisms but have mainly focused on flat filters that overlay a flat background. However, studies with flat filters have not captured all aspects of physical transparency, such as caustics, specular reflections/highlights, and shadows. Here, we investigate color matching experiments with three-dimensional transparent objects for different matching stimuli: a uniform patch and a flat filter overlaying a variegated background. Two different instructions were given to observers: change the color of the matching stimulus until it has the same color as the transparent object (for the patch and flat filter) or until it has the same color as the dye that was used to tint the transparent object (for the patch). Regardless of instruction or matching element, observers match the mean chromaticity of the glass object, but the luminance of matches depends on the backgrounds of the test image and the matching element, indicating that a color constancy-esque discounting operation is at work. We applied three models from flat filter studies to see if they generalize to our stimuli: the convergence model and the ratio of either the means (RMC) or standard deviations (RSD) of cone excitations. The convergence model does not generalize to our stimuli, but the RMC generalizes to a wider range of stimuli than the RSD. However, there is an edge case where RMC also breaks down and there may be additional features that trade-off with RMC when observers match the color of thick, curved transparent objects.

neuroscience↗

bioRxiv2018

Machine learning enables detection of early-stage colorectal cancer by whole-genome sequencing of plasma cell-free DNA

BackgroundBlood-based methods using cell-free DNA (cfDNA) are under development as an alternative to existing screening tests. However, early-stage detection of cancer using tumor-derived cfDNA has proven challenging because of the small proportion of cfDNA derived from tumor tissue in early-stage disease. A machine learning approach to discover signatures in cfDNA, potentially reflective of both tumor and non-tumor contributions, may represent a promising direction for the early detection of cancer.\n\nMethodsWhole-genome sequencing was performed on cfDNA extracted from plasma samples (N=546 colorectal cancer and 271 non-cancer controls). Reads aligning to protein-coding gene bodies were extracted, and read counts were normalized. cfDNA tumor fraction was estimated using IchorCNA. Machine learning models were trained using k-fold cross-validation and confounder-based cross-validation to assess generalization performance.\n\nResultsIn a colorectal cancer cohort heavily weighted towards early-stage cancer (80% stage I/II), we achieved a mean AUC of 0.92 (95% CI 0.91-0.93) with a mean sensitivity of 85% (95% CI 83-86%) at 85% specificity. Sensitivity generally increased with tumor stage and increasing tumor fraction. Stratification by age, sequencing batch, and institution demonstrated the impact of these confounders and provided a more accurate assessment of generalization performance.\n\nConclusionsA machine learning approach using cfDNA achieved high sensitivity and specificity in a large, predominantly early-stage, colorectal cancer cohort. The possibility of systematic technical and institution-specific biases warrants similar confounder analyses in other studies. Prospective validation of this machine learning method and evaluation of a multi-analyte approach are underway.

cancer biology↗