pyABC: distributed, likelihood-free inference

Abstract

Summary: Likelihood-free methods are often required for inference in systems biology. While Approximate Bayesian Computation (ABC) provides a theoretical solution, its practical application has often been challenging due to its high computational demands. To scale likelihood-free inference to computationally demanding stochastic models we developed pyABC: a distributed and scalable ABC-Sequential Monte Carlo (ABC-SMC) framework. It implements computation-minimizing and scalable, runtime-minimizing parallelization strategies for multi-core and distributed environments scaling to thousands of cores. The framework is accessible to non-expert users and also enables advanced users to experiment with and to custom implement many options of ABC-SMC schemes, such as acceptance threshold schedules, transition kernels and distance functions without alteration of pyABC's source code. pyABC includes a web interface to visualize ongoing and finished ABC-SMC runs and exposes an API for data querying and post-processing. Availability and Implementation: pyABC is written in Python 3 and is released under the GPLv3 license. The source code is hosted on https://github.com/neuralyzer/pyabc and the documentation on http://pyabc.readthedocs.io. It can be installed from the Python Package Index (PyPI).

Approximate Bayesian Computation with the Sliced-Wasserstein Distance

“…We compare SW-ABC against ABC using the Euclidean distance between sample variances (Euclidean-ABC), WABC with the Hilbert distance, WABC with the swapping distance and KL-ABC. Each ABC approximation was obtained using the sequential Monte Carlo sampler-based ABC method [24], which is more computationally efficient than vanilla ABC (1) and implemented in the package pyABC [25]. We provide our code in [26].…”

Efficient Exact Inference for Dynamical Systems with Noisy Measurements using Sequential Approximate Bayesian Computation

“…We implemented all the algorithms in the open-source python toolbox pyABC (https://github.com/icbdcm/pyabc, Klinger et al (2018)), which offers a state-of-the-art implementation of ABC-SMC.…”

Aggregated spatio-temporal division patterns emerge from reoccurring divisions of neural stem cells

“…From this model we simulate dividing NSCs and a BrdU-EdU-labelling experiment (Supplementary Figure 5B) and evaluate the percentage of re-dividing cells and DLS. We optimized the parameters of our model to the observed frequencies (see Methods) using approximate Bayesian computation (Klinger et al, 2018) . Our model fitted the data, in particular the plateau of re-dividing cells (Supplementary Figure 5C), and the sharp decrease of DLSs after 9h (Supplementary Figure 5D).…”

“…To infer these five parameters (scale parameter and delay of cell cycle, d cc and β cc , and S-phase, d sp and β sp , and re-division probability p re-div ) from (i) the observed re-division frequency and (ii) the fraction of DLS NSCs (see sketches in Supplementary Figure 5C,D) at every labelling interval ∆t=9h, 18h, 24h, 32h, 48h, and 72h, we apply approximate Bayesian computation (ABC) (Klinger et al, 2018) . Given the five parameters ( d cc β cc d sp β sp and p re-div ) we simulate the dynamics of 10000 dividing cells over 400-500h.…”