Skip to content

Shyam Kannan

Wet lab researcher turned computational biologist.

$ role.set('both')

Recently

The Pivot

Fifteen million possible sequences in a cyclic peptide library. The manual pipeline could realistically test a few thousand of them. I spent a lot of time at the bench in Dr. Kit Lam's lab doing exactly that, screening one bead at a time under a microscope, isolating hits by hand, confirming each one by mass spec. Over 99.9% of the sequence space never got touched.

The bottleneck was never the chemistry. It was the search. A model trained on binding data could rank a library and shortlist likely hits before anyone ever synthesized a bead. That realization is what pushed me toward computational work.

I see the same problem now at Antibodies Incorporated. Developing a therapeutic antibody means running the full experimental cycle, expression, purification, binding assays, and waiting weeks to find out if a sequence even works. A model won't replace those experiments, but it could tell us which sequences are actually worth spending bench time on.

Projects

BindScape

Predicts whether a drug and a protein will bind, using molecular fingerprints and protein language model embeddings, no 3D structure or docking involved.

+ Detail

I tested it across more than 456,000 drug-protein pairs against 495 human kinases. Adding the protein embedding actually hurt performance on new targets compared to using the fingerprint alone. The protein representation turned out to be the real bottleneck for generalizing to targets the model hadn't seen before.

0.982

AUROC, random split

0.781

AUROC, held-out target

456K+

pairs tested

BindScape results, AUROC across the random, held-out target, and scaffold splits

Antibody Sequence Landscape

Compared two protein language models on how well they separate antibody sequences by species.

+ Detail

I embedded 750 antibody heavy chain sequences across three species and tested different pooling strategies. AntiBERTy came out ahead of ESM2, and it turned out ESM2's improvement was coming entirely from how you pool the embeddings, not from masking the CDR regions like I expected going in.

+0.268

silhouette points over ESM2

750

sequences embedded

1,000

resample bootstrap

Silhouette scores comparing AntiBERTy and ESM2 across pooling strategies

TempLog

A temperature monitoring system I built and deployed on a Raspberry Pi.

+ Detail

It checks 35 to 40 lab fridges and freezers daily and logs the readings automatically, replacing close to 250 hours a year of manual logging. Built to meet FDA 21 CFR Part 11 requirements, with a tamper-evident audit trail and hardware-tied verification so the record can't quietly be altered.

~250 hrs

of manual logging saved per year

TempLog system diagram, from Raspberry Pi sensor reads through to the audit trail

Claim Jumper

In progress

Helps medical billing teams work through denied insurance claims faster.

+ Detail

It reads EOB screenshots and sorts each denial into the right next step: appeal, correction, patient bill, write off, or human review. Built solo for OpenAI Build Week 2026 using GPT-5.6 and Codex. Still in progress.

Claim Jumper flow, from EOB screenshot to routed denial outcome

More coming

Research & Industry

  1. Oct 2025 - Present

    Antibodies Incorporated

    Research Associate

    I work under Dr. Shakur Mohibi producing antibodies for academic, biotech, and pharma clients. Day to day that means Expi293 transfections, running bioreactors for hybridoma cultures, and characterizing antibodies with ELISA, Western blot, and ICC.

    ISO 13485-2016 certified, FDA-registered, cGMP/cGLP compliant.

  2. Oct 2023 - 2025

    Lam Lab, UC Davis Health

    Undergraduate Researcher

    I worked on peptide synthesis for drug delivery under Dr. Junwei Zhao, building cyclic peptide libraries with over 15 million possible sequences to find candidates that could target specific integrins. I screened them by fluorescence, pulled out the positive hits by hand, and confirmed each one with mass spec. This is where I first ran into the sequence space problem that eventually pushed me toward computational work. I also supported GMP-grade polymer synthesis for a bladder cancer clinical trial under Dr. Yanxiao Jiao, optimizing yield and purity under strict trial protocols.

  3. Apr 2022 - Nov 2022

    Enzyme Engineering Lab, UC Davis

    Undergraduate Researcher

    I engineered E. coli to produce psilocybin biosynthetic enzymes, then purified and characterized the resulting proteins using chromatography, crystallography, and mass spec.

Skills

Wet Lab

  • ELISA
  • Western blot
  • ICC/IHC
  • Cell culture
  • Bioreactor operations
  • Expi293 transfection
  • OBOC peptide synthesis
  • Fmoc chemistry
  • MALDI-MS
  • Confocal microscopy
  • GMP/GLP compliance

Computational

  • Python
  • pandas
  • scikit-learn
  • PyTorch
  • RDKit
  • ESM2
  • AntiBERTy
  • ANARCI
  • LangChain
  • Vector databases
  • Raspberry Pi deployment
  • FDA 21 CFR Part 11 compliance

Education

June 2025

University of California, Davis

B.S. Neurobiology, Physiology, and Behavior

Minor in Public Health Sciences