EXACT SPIKE TRAIN INFERENCE VIA ℓ₀ OPTIMIZATION.

Area of Science:

• Computational Neuroscience
• Neuroimaging Technology

Background:

• Advances in neuroscience allow simultaneous measurement of large neuronal populations in behaving animals.
• Fluorescence traces approximate neuronal activity, but precise spike time determination remains challenging.
• Existing methods utilize L1-penalty convex optimization for spike deconvolution.

Purpose of the Study:

• To develop a more efficient and accurate method for determining neuron spike times from fluorescence traces.
• To modify a recent L1-penalty approach by employing an L0-penalty.

Main Methods:

• A novel dynamic programming algorithm is proposed to solve the L0-penalty optimization problem.
• The algorithm efficiently finds the global optimum for spike deconvolution.
• An R-language implementation (LZeroSpikeInference package) was developed and tested.

Main Results:

• The L0-penalty optimization problem, contrary to conventional assumptions, is shown to be computationally tractable.
• The proposed dynamic programming algorithm achieves substantial improvements over the L1-penalty method.
• The R-language implementation processes large fluorescence traces (100,000 timesteps) in minutes.

Conclusions:

• The L0-penalty approach offers a significant advancement in accurately inferring neural spiking activity from fluorescence imaging data.
• The developed algorithm provides an efficient and effective solution for spike deconvolution in computational neuroscience.
• Open-source software is available for broader application and validation.