Design of complex neuroscience experiments using mixed-integer linear programming

Over the past few decades, neuroscience experiments have become increasingly complex and naturalistic. Experimental design has in turn more challenging, as must conform to an ever-increasing diversity of constraints. In this article, we demonstrate how process can be greatly assisted using optimization tool known mixed-integer linear programming (MILP). MILP provides a rich framework for incorporating many types real-world constraints into experiment. We introduce mathematical foundations MILP, compare other experimental techniques, provide four case studies used solve challenges. Many different tools been designing experiments, including optimal designs (Dale, 1999Dale A.M. Optimal event-related fMRI.Hum. Brain Mapp. 1999; 8: 109-114Crossref PubMed Scopus (1339) Google Scholar; Wager Nichols, 2003Wager T.D. Nichols T.E. Optimization fMRI: general genetic algorithm.Neuroimage. 2003; 18: 293-309Crossref (328) Scholar) large variety combinatorial (Friston et al., 1997Friston K.J. Buechel C. Fink G.R. Morris J. Rolls E. Dolan R.J. Psychophysiological modulatory interactions neuroimaging.Neuroimage. 1997; 6: 218-229Crossref (2352) Scholar, Friston 1998Friston Josephs O. Rees G. Turner R. Nonlinear responses fMRI.Magn. Reson. Med. 1998; 39: 41-52Crossref (532) Henson, 2007Henson Efficient fMRI.in: Penny W.D. Ashburner J.T. Kiebel S.J. eds, Statistical Parametric Mapping: The Analysis Functional Images. Academic, 2007: 193-210Crossref (99) Scholar). These work exceedingly well their intended use cases, but they are unable meet challenges that arise neuroimaging particular, idiosyncratic reflect structure real world. Conventional usually specialized accommodate very limited set constraints, little flexibility integrating additional design. A flexible approach is explicitly frame constrained problem. Given researcher wants create experiment optimally tests some scientific hypotheses. This framing by technique suited representing experiments. already fields such hydrology (Hsu Yeh, 1989Hsu N.-S. Yeh W.W.-G. Optimum Design Parameter Identification Groundwater Hydrology.Water Resour. Res. 1989; (Published online July 9, 2010)https://doi.org/10.1029/WR025i005p01025Crossref (51) Scholar), genetics (McClosky Tanksley, 2013McClosky B. Tanksley S.D. Optimizing Genetics.J. Optim. Theory Appl. 2013; 157: 520-532Crossref (2) agricultural economics (Chou, 1961Chou T.-H. Mixed Integer Programming with Application Dynamic Non-Linear Models Agriculture. Dissertation (Iowa State University).1961Google educational assessment (Linden 2004Linden W.J.v.d. Veldkamp B.P. Carlson J.E. Balanced Incomplete Block Designs Educational Assessments.Appl. Psychol. Meas. 2004; September 1, 2004)https://doi.org/10.1177/0146621604264870Crossref (28) also context phase unwrap fMRI data (Jenkinson, 2003Jenkinson M. Fast, automated, N-dimensional phase-unwrapping algorithm.Magn. 49: 193-197Crossref (522) sampling prescription diffusion imaging (Cheng 2014Cheng Shen D. Yap P.-T. Designing single- multiple-shell schemes MRI spherical code.in: Golland P. Hata N. Barillot Hornegger Howe Medical Image Computing Computer-Assisted Intervention – MICCAI 2014. Lecture Notes Computer Science, vol 8675. Springer, 2014: 281-288Crossref (13) analyze resting-state (Demertzi 2014Demertzi A. Gómez F. Crone J.S. Vanhaudenhuyse Tshibanda L. Noirhomme Q. Thonnard Charland-Verville V. Kirsch Laureys S. Soddu Multiple system-level baseline connectivity disrupted patients consciousness alterations.Cortex. 2014; 52: 35-46Crossref (134) Costa 2015Costa Smith T. Cussens Duff E.P. Makin T.R. Searching Multiregression Resting-State Networks Using Programming.Bayesian Anal. 2015; 10: 441-470Crossref (19) Henry Gates, 2017Henry Gates K. Causal Search Procedures Review Suggestions.Behaviormetrika. 2017; 44: 193-225Crossref (14) 2019Costa J.Q. group analysis networked time series.J. Stat. Plan. Inference. 2019; 198: 43-61Crossref (3) perform source localization MEG (Freschi, 2010Freschi Localization Sources Activity: Approach.IEEE Trans. Magn. 2010; 29: 833-843Google However, identify any functional article develops demonstrates conceptual information provided applicable Challenges describes associated naturalistic Mathematical introduction mathematics MILP. explores where uniquely equipped optimize designs. Lastly, Practical considerations compares techniques discusses its limitations. Although it often desirable simple possible, functions brain only investigated because responds qualitatively manner when stimuli or tasks overly simplified unnatural (Wu 2006Wu M.C. David S.V. Gallant J.L. Complete characterization sensory neurons system identification.Annu. Rev. Neurosci. 2006; 477-505Crossref (202) Oversimplification undermine ecological validity generalizability situations (Schmuckler, 2001Schmuckler M.A. What Is Ecological Validity? Dimensional Analysis.Infancy. 2001; 2: 419-436Crossref (228) Wu Sonkusare 2019Sonkusare Breakspear Guo Naturalistic Stimuli Neuroscience: Critically Acclaimed.Trends Cogn. Sci. 23: 699-714Abstract Full Text PDF (125) Qualitative differences evoked versus demonstrated subsystems, vision (Dan Poo, 2004Dan Y. Poo M.-M. Spike timing-dependent plasticity neural circuits.Neuron. 23-30Abstract (689) David, 2004David Natural Stimulus Statistics Alter Receptive Field Structure V1 Neurons.J. 24: 6991-7006Crossref audition (Rieke 1995Rieke Bodnar D.A. Bialek W. Increase Rate Efficiency Information Transmission Primary Auditory Afferents.Proc. Biol. 1995; 262: 259-265Crossref Woolley 2005Woolley S.M.N. Fremouw Hsu Theunissen F.E. Tuning spectro-temporal modulations mechanism auditory discrimination natural sounds.Nat. 2005; 1371-1379Crossref (204) Wild 2017Wild C.J. Linke A.C. Zubiaurre-Elorza Herzmann Duffy H. Han V.K. Lee D.S.C. Cusack Adult-like Processing Sounds Cortex 3- 9-Month Old Infants.NeuroImage. 623-634Crossref (15) language comprehension (Vandenberghe 2002Vandenberghe Nobre Price response left temporal cortex sentences.J. 2002; 14: 550-560Crossref (276) Huth 2016Huth A.G. de Heer W.A. Griffiths T.L. Speech Reveals Semantic Maps That Tile Human Cerebral Cortex.Nature. 2016; 532: 453-458Crossref (610) face perception (Allison 2000Allison Puce McCarthy Social from visual cues: role STS region.Trends 2000; 4: 267-278Abstract (1800) Kilgour 2005Kilgour A.R. Kitada Servos James T.W. Lederman Haptic identification activates ventral occipital areas: study.Brain 59: 246-257Crossref (33) Gauthier 2009Gauthier I. Klaiman Schultz R.T. Face composite effects reveal abnormal processing Autism spectrum disorders.Vision 2009; 470-478Crossref (59) social cognition (Zaki Ochsner, 2009Zaki Ochsner need cognitive cognition.Ann. NY Acad. 1167: 16-30Crossref (154) Hari 2015Hari Henriksson Malinen Parkkonen Centrality Interaction Function.Neuron. 88: 181-193Abstract (159) Babiloni Astolfi, 2014Babiloni Astolfi hyperscanning techniques: past, present future.Neurosci. Biobehav. 76-93Crossref (263) memory (Nielson 2015Nielson D.M. T.A. Sreekumar Dennis Sederberg P.B. Hippocampus Represents Space Time during Retrieval Real-World Memories.Proc. Natl. USA. 112: 11078-11083Crossref (111) Ren 2018Ren Nguyen V.T. Lv Pang Eickhoff S.B. C.C. Effective anterior hippocampus predicts recollection confidence retrieval.Nat. Commun. 2018; 9: 4875Crossref (20) Kauttonen 2018Kauttonen Hlushchuk Jääskeläinen I.P. Tikka mechanisms underlying cue-based memorizing free viewing movie Memento.Neuroimage. 172: 313-325Crossref (17) Detailed discussions most appropriate found Schmuckler, Scholar Scholar. To avoid problems related oversimplification, incorporate world conventional not expressive enough represent structure. include designs, broad category includes m-sequences (Sutter, 1987Sutter E.E. Nonstochastic Approach Time-Domain Analysis..Advanced Methods Physiological System Modelling, V.Z. Marmarelis, ed.1. Biomed. Simul. Univ. Calif., Los Angeles, CA1987: 303-315Google Benardete Victor, 1994Benardete E.A. Victor J.D. An extension m-sequence multi-input nonlinear systems.in: Marmarelis Advanced Modeling. 1994: 87-110Crossref Buračas Boynton, 2002Buračas G.T. Boynton G.M. M-sequences.Neuroimage. 16: 801-813Crossref factorial Hall 2000Hall Haggard M.P. Akeroyd Summerfield A.Q. Palmer Elliott M.R. Bowtell R.W. Modulation task measured 107-119Crossref (100) Gurd 2002Gurd J.M. Amunts Weiss P.H. Zafiris Zilles Marshall J.C. Posterior parietal implicated continuous switching between verbal fluency tasks: study clinical implications.Brain. 125: 1024-1038Crossref (137) latin squares Hagberg 2001Hagberg G.E. Zito Patria Sanes J.N. Improved detection signals probability functions.Neuroimage. 1193-1205Crossref (79) others. Each these useful selecting arranging stimuli, tasks, conditions grow complexity, difficult impossible use. section, will describe One common motif number variables Thus, requires decisions about selection arrangement variables. For example, utilize hundreds thousands (Thorpe 1996Thorpe Fize Marlot Speed human system.Nature. 1996; 381: 520-522Crossref (2541) Kay 2008Kay K.N. Naselaris Prenger Identifying images activity.Nature. 2008; 452: 352-355Crossref (791) Cognitive control might require subjects learn tens novel mappings task-relevant (Gauthier 1998Gauthier Williams Tarr M.J. Tanaka Training ‘greeble’ experts: studying expert object recognition processes.Vision 38: 2401-2428Crossref (320) Badre 2010Badre Kayser A.S. D’Esposito Frontal discovery abstract action rules.Neuron. 66: 315-326Abstract (200) Dresler 2017Dresler Shirer W.R. Konrad B.N. Müller N.C.J. Wagner I.C. Fernández Czisch Greicius M.D. Mnemonic Reshapes Support Superior Memory.Neuron. 93: 1227-1235Abstract (86) each required make select arrange conditions. decision depend on speech contain unique features affect resultant activity (Nishimoto 2011Nishimoto Vu A.T. Benjamini Yu Reconstructing experiences movies.Curr. 2011; 21: 1641-1646Abstract 2012Huth Nishimoto semantic space representation categories across brain.Neuron. 2012; 76: 1210-1224Abstract (464) Sudre 2012Sudre Pomerleau Palatucci Wehbe Fyshe Salmelin Mitchell Tracking coding perceptual concrete nouns.Neuroimage. 62: 451-463Crossref (83) 2017de Hierarchical Cortical Organization Processing.J. 37: 6539-6557Crossref (90) Even synthetic dot-motion relevant parameters, color, color coherence, motion direction, speed, dot count, size, cluster fixation position (Newsome Paré, 1988Newsome W.T. Paré E.B. selective impairment following lesions middle area (MT).J. 1988; 2201-2211Crossref Mante 2013Mante Sussillo Shenoy K.V. Newsome Context-dependent computation recurrent dynamics prefrontal cortex.Nature. 503: 78-84Crossref (709) Optimally thus accounting features. do adequately scale dimensions low-dimensional impose undesirable length way which combine (Vieira 2011Vieira Sanchez Kienitz K.H. Mischel C.N.B. Generating Improving Orthogonal Programming.Eur. Oper. 215: 629-638Crossref (22) difficulties worse grows complexity Tools fractional balanced incomplete block developed better scalability high dimensionality (Montgomery, 2008Montgomery D.C. Experiments. Wiley, 2008Google still relatively small numbers dimensions. ideal would flexibly whatever Another challenge individual dimension own words, (related variables) characteristics special treatment. consider composed binary contrasts, those localize cortical regions interest like Fusiform Area Parahippocampal Place (Saxe 2006Saxe Brett Kanwisher Divide Conquer: Defense Localizers Neuroimage.30. 2006: 1088-1096Google type experiment, condition two possible levels, off. able precede follow condition. treated identically Now executive subject navigates tree. Decision trees gambling (Paliwal 2014Paliwal Petzschner F.H. Schmitz A.K. Tittgemeyer Stephan K.E. model-based impulsivity slot-machine paradigm.Front. Hum. 428Crossref foraging (Kolling 2012Kolling Behrens T.E.J. Mars R.B. Rushworth M.F.S. Neural foraging.Science. 336: 95-98Crossref (372) (Daw 2005Daw N.D. Niv Dayan Uncertainty-based competition dorsolateral striatal systems behavioral control.Nat. 1704-1711Crossref (1444) Gläscher 2010Gläscher Daw O’Doherty J.P. States rewards: dissociable prediction error model-free reinforcement learning.Neuron. 585-595Abstract (694) Here, series decisions, specific consequences states follow. Unlike contrast subsequent rewards, parameters. example navigation Consider instructed navigate locations map (Hartley 2003Hartley Maguire Spiers H.J. Burgess well-worn route path less traveled: distinct bases wayfinding humans.Neuron. 877-888Abstract (612) Maguire, 2006Spiers Spontaneous mentalizing interactive task: study.Neuropsychologia. 1674-1682Crossref Depending hypotheses being tested, important account map, particular destinations, landmarks, routes. naturally represented spatial graph data. neither nor directly tools. When identically, straightforward struggle tree manipulate structures There practice pre-existing media done movies books (Wehbe 2014Wehbe Murphy Talukdar Ramdas Simultaneously uncovering patterns involved story reading subprocesses.PLoS ONE. e112575Crossref (116) music (Khorram-Sefat 1997Khorram-Sefat Dierks Herminghaus Hacker Correlates Music Listening Cortical-Cerebellar Activation Pattern.Psychiatry Neuroimaging. 68: 167-168Google podcasts (Huth maps video games (Mathiak Weber, 2006Mathiak Weber Toward correlates behavior: violent games.Hum. 27: 948-956Crossref (142) enable were originally created purpose, limitations dictate used. exemplars phenomena studied, confounds prevent certain tested. available materials significant considered throughout process. Most materials. Instead, separately verify produced compatible available. becomes iterative adjusting parameters checking whether both feasible given assess full range so optimally. Finally, mention single highly diverse interdependent. exist beyond mentioned far. almost all restrictions total amount collected due endurance, patient availability, research budgets. tested constrain high-level semantics low-level luminance help ensure observed caused (Dell’Acqua Grainger, 1999Dell’Acqua Grainger Unconscious priming pictures.Cognition. 73: B1-B15Crossref (126) seek properties system, hemodynamics spatiotemporal resolution modality (Amaro Barker, 2006Amaro Jr., Barker G.J. Study basic principles.Brain 60: 220-232Crossref (314) Integrating challenging interact conflict one another. constraining timing