Revealing unexpected complex encoding but simple decoding mechanisms in motor cortex via separating behaviorally relevant neural signals
Abstract
In motor cortex, behaviorally relevant neural responses are entangled with irrelevant signals, which complicates the study of encoding and decoding mechanisms. It remains unclear whether behaviorally irrelevant signals could conceal some critical truth. One solution is to accurately separate behaviorally relevant and irrelevant signals at both single-neuron and single-trial levels, but this approach remains elusive due to the unknown ground truth of behaviorally relevant signals. Therefore, we propose a framework to define, extract, and validate behaviorally relevant signals. Analyzing separated signals in three monkeys performing different reaching tasks, we found neural responses previously considered to contain little information actually encode rich behavioral information in complex nonlinear ways. These responses are critical for neuronal redundancy and reveal movement behaviors occupy a higher-dimensional neural space than previously expected. Surprisingly, when incorporating often-ignored neural dimensions, behaviorally relevant signals can be decoded linearly with comparable performance to nonlinear decoding, suggesting linear readout may be performed in motor cortex. Our findings prompt that separating behaviorally relevant signals may help uncover more hidden cortical mechanisms.
Type
Publication
eLife
🔶 Understanding Neural Encoding and Decoding in Motor Cortex 🔶
🎯 Research Goal
Understanding how motor cortex (MC) encodes and decodes movement behaviors is a fundamental goal of neuroscience
(Kriegeskorte & Douglas, 2019; Saxena & Cunningham, 2019)
- Behavior = Measured task-specific behavioral variables (e.g., arm kinematics)
- Terms like “behaviorally relevant” or “irrelevant” refer only to these variables
⚠️ Key Challenge
❗ Neural signals are entangled:
- Relevant neural responses are mixed with:
- Irrelevant variables (Fusi et al., 2016; Rigotti et al., 2013)
- Ongoing neural noise (Azouz & Gray, 1999; Faisal et al., 2008)
🔍 This raises a critical question:
Could irrelevant signals conceal important facts about neural encoding and decoding?
🧩 Implications of Irrelevant Signals
🔬 Neural Encoding
- Irrelevant signals may:
- Mask small neural components
- Lead to underestimation of weakly tuned neurons
- Obscure informative low-variance PCs
Example Misconceptions:
- Weakly tuned neurons often ignored (Georgopoulos et al., 1986; Hochberg et al., 2012)
- Low-variance PCs discarded as noise (Churchland et al., 2012; Gallego et al., 2018, 2020)
❓ Do these components truly lack information, or are they just obscured?
🧠 Neural Decoding
- Irrelevant signals make information readout harder (Pitkow et al., 2015; Yang et al., 2021)
- Debate: Is motor cortex decoding:
- Linear (plausible and common)
- Nonlinear (performs better in practice, e.g., Glaser et al., 2020)
❓ Are irrelevant signals the reason nonlinear models outperform linear ones?
🧰 Proposed Solution: Signal Separation Framework
✅ Goal:
Separate behaviorally relevant and irrelevant signals at:
- Single-neuron
- Single-trial levels
🚧 Challenges:
- Ground truth of relevant signals is unknown
- Existing methods rely on assumptions:
- Focus on latent population (not single neurons)
- Assume specific dynamics (linear/nonlinear)
Limitations of Existing Approaches:
- Trial-averaged signals lose single-trial dynamics
- Assumption-based models miss out on nonconforming activity
🆕 Our Novel Framework
✨ Key Idea:
Define, extract, and validate behaviorally relevant signals based on:
- Encoding: Should resemble raw signals (preserve neural features)
- Decoding: Should retain maximal behavioral information
➡️ Establishes foundation for accurate mechanistic analysis
🧪 Experimental Setup
- Subjects: Three monkeys
- Task: Different reaching tasks
- Behavioral variable: Movement kinematics
🔍 Key Findings
1. Behaviorally Irrelevant Signals
- Explain a large portion of trial-to-trial variability
- Evenly distributed across dimensions of relevant signals
2. Neural Encoding Insights
- Irrelevant signals obscure behaviorally meaningful activity
- Responses with nonlinear encoding are especially affected
🚫 Previously Ignored Signals:
- Weakly tuned neurons
- Low-variance PCs
✅ Actually encode rich behavioral information in nonlinear ways
- Reveal neuronal redundancy
- Suggest a higher-dimensional neural representation of movement
🧠 Synergistic Effect:
- Combining small & large variance PCs improves decoding
- Small variance PCs enhance fine-grained speed control → Crucial for precise motor control
3. Neural Decoding Insights
- Irrelevant signals complicate decoding
- Linear decoders can match nonlinear decoders once irrelevant signals are removed → Supports the possibility of linear readout in motor cortex
✅ Encoding = Complex
✅ Decoding = Simple
🚀 Broader Impact
🔧 Implications for:
- Brain-machine interfaces (BMIs): More accurate & robust decoding
- Systems neuroscience: General framework to separate signals and uncover hidden neural mechanisms across cortical areas
📚 Key References
- Kriegeskorte & Douglas, 2019
- Saxena & Cunningham, 2019
- Fusi et al., 2016; Rigotti et al., 2013
- Azouz & Gray, 1999; Faisal et al., 2008
- Georgopoulos et al., 1986; Hochberg et al., 2012; Wodlinger et al., 2015
- Churchland et al., 2012; Gallego et al., 2018, 2020
- Pitkow et al., 2015; Yang et al., 2021
- Glaser et al., 2020; Willsey et al., 2022
- Sani et al., 2021; Hurwitz, 2021; Zhou, 2020
- Kobak et al., 2016; Rouse & Schieber, 2018
