mlf-core: a framework for deterministic machine learning.

MOTIVATION: Machine learning has shown extensive growth in recent years and is now routinely applied to sensitive areas. To allow appropriate verification of predictive models before deployment, models must be deterministic. Solely fixing all random seeds is not sufficient for deterministic machine learning, as major machine learning libraries default to the usage of nondeterministic algorithms based on atomic operations. RESULTS: Various machine learning libraries released deterministic counterparts to the nondeterministic algorithms. We evaluated the effect of these algorithms on determinism and runtime. Based on these results, we formulated a set of requirements for deterministic machine learning and developed a new software solution, the mlf-core ecosystem, which aids machine learning projects to meet and keep these requirements. We applied mlf-core to develop deterministic models in various biomedical fields including a single-cell autoencoder with TensorFlow, a PyTorch-based U-Net model for liver-tumor segmentation in computed tomography scans, and a liver cancer classifier based on gene expression profiles with XGBoost. AVAILABILITY AND IMPLEMENTATION: The complete data together with the implementations of the mlf-core ecosystem and use case models are available at https://github.com/mlf-core.

Transcriptional behavior of the HIV-1 promoter in context of the BACH2 prominent proviral integration gene.

Chronic infection with human immunodeficiency virus (HIV)-1 is characterized by accumulation of proviral sequences in the genome of target cells. Integration of viral DNA in patients on long-term antiretroviral therapy selectively persists at preferential loci, suggesting site-specific crosstalk of viral sequences and human genes. This crosstalk likely contributes to chronic HIV disease through modulation of host immune pathways and emergence of clonal infected cell populations. To systematically interrogate such effects, we undertook genome engineering to generate Jurkat cell models that replicate integration of HIV-1 long terminal repeat (LTR) sequences at the BTB and CNC Homolog 2 (BACH2) integration locus. This locus is a prominent HIV-1 integration gene in chronic infection, found in 30 % of long-term treated patients with mapped proviral integrations. Using five clonal models carrying an LTR-driven reporter at different BACH2 intergenic regions, we here show that LTR transcriptional activity is repressed in BACH2 regions associated with proviral-DNA integrations in vivo but not in a control region. Our data indicates that this repression is in part epigenetically regulated, particularly through DNA methylation. Importantly, we demonstrate that transcriptional activity of the LTR is independent of BACH2 gene transcription and vice versa in our models. This suggests no transcriptional interference of endogenous and HIV-1 promoters. Taken together, our study provides first insights into how activity of HIV-1 LTR sequences is regulated at the BACH2 locus as prominent example for a recurrently-detected integration gene in chronic infection. Given the importance of integration-site dependent virus/host crosstalk for chronic HIV disease, our findings for the BACH2 locus have potential implications for future therapeutic strategies.