RT @DynamicsSIAM: "How causal perspectives can inform problems in computational neuroscience" (by Eric W. Bridgeford, Brian S. Caffo, Maya….

Let's improve the reliability and validity of neuroimaging research together! Be sure to check out our new complementary software package 11/11.

The implications? HUGE, especially as new methods are rolled out for batch effect correction leveraging deep learning, which are heavily susceptible to covariate distribution shift. Ignoring causality yields bias in your models. #deeplearning #ai #biasinai 10/11.

Our mega analysis (1700+ connectomes) shows how causal methods reveal the true extent of batch effects, while traditional methods falter. #realdata #results 9/11

Causal ComBat: Supercharging ComBat with causal reasoning. Removes batch effects without sacrificing valuable biological variation. #biostatistics #combat 8/11

Causal DCorr: a non-parametric weapon for detecting batch effects in non-Euclidean data like fMRI connectomes. #neuroimaging #connectomics 7/11.

We present a unified, causal framework. New methods for detection, estimation, and correction that outperform existing techniques under confounding. #statistics #datascience 6/11.

Traditional methods often fail under these types of confounding. They often fail to distinguish batch effects from other sources of variation. #bias #methodology 5/11

Multi-site studies? Awesome for diverse samples! But confounding is inevitable. Different sites typically mean samples with very different demographics, making isolating batch effects from demographic signals difficult. #experimentaldesign 4/11

Big idea: Batch effects are causal effects. How you collect and process data directly impacts your measurements of biological phenomena. 3/11.

The reproducibility crisis keeps knocking. One major culprit? Batch effects masking true signals and creating spurious findings. Let's fix this! #reproducibility #datascience #neuroimaging 2/12.

What are batch effects? How do they wreak havoc on our multi-site (neuroimaging) studies? 🤔 And where are current approaches falling short? Time for some causal clarity! #causality with @neuro_data @g_kiar @MilhamMichael @ImagingNeurosci 1/11.

RT @neuro_data: When no answer is better than a wrong answer: a causal perspective on batch effects. I still love….

We are excited to hear how you agree, or disagree, with the framework of batch effects we have come up with, and how we can work together to continue to improve! 11/11.

the limitations of the actual data we are obtaining in statistical connectomics (and other biomedical datasets), and how those limitations can be overcome or better understood 10/11.

To me, the questions we raise here motivate the importance of stepping back from continuously looking for new downstream biological inference tasks to study, and focusing much future effort on better understanding 9/11.

We demonstrate that existing strategies fail to differentiate between veridical biological signal and sources of variability due to other experimental design factors on a large neuroimaging dataset from the Consortium for Reliability and Reproducibility 8/11

We augment the ComBat procedure with causal reasoning to obtain Causal ComBat, which uses causal methods to limit the removal of variation due to observed demographic variables 7/11.

We develop a non-parametric technique, Causal DCorr, to estimate and detect the presence of batch effects in non-Euclidean data such as fMRI connectomes 6/11

Unlike many approaches which focus only on mitigation of batch effects, we first propose a formal definition of a causal batch effect, revealing the limitations of assumptions which are implicit in many existing approaches 5/11